Early Right Ventricular Assist Device Utilization in Patients Undergoing Continuous-Flow Left Ventricular Assist Device Implantation: Incidence and Risk Factors from INTERMACS

Abstract

Background

To investigate pre-implant risk factors associated with early right ventricular assist device (RVAD) use in patients undergoing continuous-flow left ventricular assist device (CF-LVAD) surgery.

Methods and Results

Patients in the INTERMACS registry who underwent primary CF-LVAD surgery were examined for concurrent or subsequent RVAD implantation within 14 days of LVAD. Risk factors for RVAD implantation and the combined endpoint of RVAD or death within 14 days of LVAD were assessed with stepwise logistic regression. We compared survival between patients with and without RVAD using Kaplan-Meier method and Cox proportional hazards modeling. Of 9,976 patients undergoing CF-LVAD implantation, 386 patients (3.9%) required an RVAD within 14 days of LVAD surgery. Pre-implant characteristics associated with RVAD use included INTERMACS patient profiles 1 and 2, the need for preoperative ECMO or renal replacement therapy, severe pre-implant tricuspid regurgitation, history of prior cardiac surgery, and concomitant procedures other than TV repair at the time of LVAD. Hemodynamic determinants included elevated right atrial pressure, reduced pulmonary artery pulse pressure, and reduced stroke volume. The final model demonstrated good performance for both RVAD implant (AUC 0.78) and the combined end-point of RVAD or death within 14 days (AUC 0.73). Compared to patients receiving an isolated LVAD, patients requiring RVAD had decreased 1- and 6-month survival: 78.1% versus 95.8% and 63.6% versus 87.9% respectively (p<0.0001 for both).

Conclusions

The need for RVAD implantation after LVAD is associated with indices of global illness severity, markers of end-organ dysfunction and profiles of hemodynamic instability.

Left ventricular assist devices (LVADs) are increasingly used to treat refractory heart failure (HF) not only as a bridge to transplant (BTT) but also as destination therapy¹. The development of severe right heart failure (RHF) following LVAD surgery is a serious complication associated with increased morbidity and mortality². Patients with post-operative RHF suffer from greater debilitation, delayed restoration of end-organ function, prolonged intensive care unit and hospital length of stay, and decreased survival^2–5.

Therapeutic interventions for severe RHF following LVAD implantation are limited and include inotropic drugs or biventricular mechanical circulatory support (BiV-MCS). Planned temporary right ventricle (RV) support is feasible and has been demonstrated to improve outcomes compared to rescue therapy^{4, 5}. Improved discrimination regarding risk of RHF would better inform clinical decision making for patients being considered for LVAD therapy. Currently, there is no consensus on how to best define a target population for isolated left-ventricular support. A number of RHF prediction rules have been previously developed^{2, 6–10}. However, available risk scores have been typically derived from single-center cohorts, employed variable definitions of RHF, identified inconsistent predictors^{11, 12}, and have demonstrated modest predictive value when validated in independent cohorts¹². The goal of the this study was to further explore pre-implant characteristics associated with right ventricular assist device (RVAD) use in patients undergoing continuous-flow (CF) LVAD surgery utilizing data from the multicenter Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS).

Methods

Patient Population and Characteristics

The Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) is a prospective registry of approved durable mechanical circulatory support devices implanted in the North America (http://www.INTERMACS.org). The INTERMACS protocol was approved by the National Institutes of Health, the Institutional Review Board (IRB) at the Data Coordinating Center at the University of Alabama at Birmingham, and each of the IRBs of the 144 participating hospitals. The study cohort includes patients ≥ 19 years of age who received a primary, intracorporeal continuous-flow (CF) LVAD between June 2006 and March 2014.

Baseline epidemiological, biochemical, echocardiographic, and hemodynamic characteristics were obtained for all patients. LVAD type (axial versus centrifugal flow) was not available in the dataset for comparison. Concomitant surgical procedures performed during the LVAD implant were included as they typically represent planned interventions and are potentially modifiable risk factors. Baseline oral medications were not included as candidate variables as both warfarin and INR were selected in an early model. A previously defined model of RHF was examined². From this model a HeartMate II RV Risk Score was created: right atrial pressure / pulmonary capillary wedge pressure (RAP/PCWP) > 0.63 – 2 points; blood urea nitrogen (BUN) level > 39 mg/dl – 2 points; pre-operative mechanical ventilation – 6 points¹³. This was evaluated as a dichotomous variable with a score > 2 points (high risk) or ≤ 2 points (low risk)¹⁴.

The primary outcome was implantation of an RVAD within 14 days of the index LVAD implant. Coding within the INTERMACS dataset did not allow further discrimination of RVAD devices as concomitant versus delayed or temporary versus permanent. Many early deaths after LVAD are attributable to severe RHF, so we performed a separate analysis to assess the model’s ability to discriminate the risk of a composite end-point of death or RVAD within 2 weeks. Odds ratios for the composite end-point model were re-estimated and the changes in beta-coefficients between the models were compared.

Statistical Analysis

Risk Model

Categorical variables are expressed as frequencies and percentages. Continuous variables are expressed as means if normally distributed or as medians if skewed. Differences between groups regarding categorical variables were compared using the X² test, and differences between groups regarding continuous variables were compared with the Student’s t-test or non-parametric Wilcoxon’s rank sum test as appropriate.

To identify risk factors for RVAD implantation, univariate logistic regression models were generated from all candidate variables. Variables missing > 50% of data were excluded. For variables with < 50% missing data, ten imputed datasets were created using SAS 9.4 PROC MI procedure using the Markov Chain Monte Carlo (MCMC) method. All variables screened initially for inclusion the in the model building process were used for the imputation models. Extreme values at the tails of the distribution for each variable were truncated at the 1^st and 99^th percentiles. Values beyond the 1^st and 99^th percentiles were replaced with values of the 1^st and 99^th percentiles respectively. Derived variables were calculated using the imputed data set. To begin the model building process, stepwise forward logistic regression was performed using a threshold of P < 0.05 to enter and stay in the model. Given that ten copies of the data set were generated, each subject was assigned a weight of 1/10. Variables from the final regression model were run in each of the ten copies of the dataset and the results were pooled using the SAS procedure PROC MIANALYZE to generate corrected p-values and parameter estimates.

Non-adjusted associations of continuous variables with outcomes were assessed for linear fit using restricted cubic splines. When the assumption of a linear relationship between a continuous variable and log odds of the outcome violated, variables were truncated to an inflection point occurring at a change in slope and loss of linearity beyond an extreme value (Supplemental Figure 1).

A number of candidate variables correlated with each other. The choice of which variable to include in the model was made based both on clinical insight as well as statistical measures. When clinically there was no preference of which of two strongly correlated variables to include in the model, we allowed the one with the strongest statistical association to be included. All model refinements based on both clinical insight and statistical significance were run on the 10 imputed datasets with pooled parameters estimated with proc MIANALYZE account for both within and between dataset variability. Parameter estimates from the final model were used to get fitted or predicted values of risk RVAD for each subject in the 10 copies of the imputed data. These fitted values were first used to get an overall ROC area across all copies of the data. Patients were then grouped based on predicted risk of RVAD from the final model (< 1%; 1 ≥ risk < 5%, 5 ≥ risk < 10%; and > 10%) and baseline characteristics were compared across these risk strata.

Survival

Survival analyses were performed using the Kaplan-Meier method, with censoring for heart transplantation or LVAD explant for cardiac recovery. The log-rank test was used to compare survival between patients with and without RVAD. Cox proportional hazard modeling was performed on the imputed dataset. The model was run for each of the 10 datasets and the estimates were pooled to determine the adjusted hazard ratios (HR) and 95% confidence intervals (CI). Adjustment variables in the Cox PH survival model included those from the final RVAD model in addition to age, gender, serum sodium, and blood urea nitrogen (BUN). A global test of proportional hazards was performed using the global correlation test on weighted Schoenfeld residuals. All statistical analyses were performed with SAS version 9.4 (SAS institute Inc., Cary, NC).

Results

The final analysis cohort consisted of 9,976 patients undergoing primary CF-LVAD, and 386 (3.9%) received an RVAD within 14 days of LVAD implantation (Figure 1). Patients requiring an RVAD were generally more ill at the time of LVAD surgery as represented by a greater proportion with INTERMACS profiles 1 or 2, and NYHA Class IV functional status (Table 1). Patients requiring RVAD were less likely to be receiving neurohormonal antagonist therapy and had a worse biochemical profile. RVAD recipients had more severe hemodynamic derangements including higher right atrial pressure (RAP), lower pulmonary artery pulse pressure (PAPP), and lower stroke volume (SV). Patients requiring RVAD placement also had substantially higher pre-implant rate of temporary mechanical support including ECMO and IABP, mechanical ventilation, and renal replacement therapy. The HeartMate II RV Risk Score was ≥ 2 in 77% of patients requiring RVAD versus 53% of patients receiving an LVAD alone (p<0.0001).

Figure 1.

Consort diagram of the study population from the INTERMACS registry. Patients requiring a right ventricular assist device (RVAD) use within 14 days of continuous flow left ventricular assist device implantation were compared to those who did not.

TABLE 1.

Baseline Characteristics

Variable	All (N=9976)	Early RVAD (N=386)	No Early RVAD (N=9590)	p-value
Demographics
Age ≥ 60 years	4722 (47.3%)	161 (41.7%)	4561(47.6%)	0.02
Male	7870 (78.9%)	303 (78.5%)	7567 (78.9%)	0.84
Destination Therapy	3668 (36.8%)	121 (31.3%)	2547 (37.0%)	0.02
INTERMACS Profile				<.0001
1 (Critical Cardiogenic)	1493 (15.0%)	153 (39.6%)	1340 (14.0%)
2 (Progressive Decline)	3805(38.1%)	155 (40.2%)	3650(38.1%)
3 (Stable on inotrope)	2791 (28.0%)	47 (12.2%)	2744 (28.6%)
≥ 4	1887 (18.9%)	31 (8.0%)	1856 (19.4%)
Medical History
Ischemic etiology	4748 (48.0%)	189 (49.7%)	4559 (47.9%)	0.48
Prior CABG or valve surgery	2808 (28.1%)	134 (34.7%)	2674 (27.9%)	0.003
Ascites	595(6.6%)	38 (11.1%)	557 (6.4%)	0.0006
Baseline Medications
ACEI or ARB	3758 (39.3%)	85 (23.0%)	3673 (40.0%)	<.0001
Aldosterone Antagonist	3958 (42.2%)	113 (31.0%)	3845 (42.6%)	<.0001
Amiodarone	3822 (40.4%)	171 (46.3%)	3651 (40.1%)	0.02
Beta Blockers	5463 (56.5%)	1636 (36.7%)	5327 (57.2%)	<.0001
Loop Diuretics	8660 (87.2%)	313 (81.1%)	8347 (87.4%)	0.0003
Laboratory Variables
Sodium (mmol/L)	135 (110-160)	134 (110-153)	135 (110-160)	0.002
Creatinine (mg/dL)	1.3 (0.01-15)	1.4 (0.2-6.9)	1.3 (0.01-15)	0.0005
BUN (mg/dL)	25 (1-285.71)	29 (4-160)	25 (1-285.71)	<.0001
AST (u/L)	30 (3-10,000)	44.5 (11-10,000)	29(3-6713)	<.0001
Albumin (g/dL)	3.4 (0-8)	3.2 (0.27-6)	3.4 (0-8)	<.0001
Total Bilirubin (mg/dL)	1 (0.08-50)	1.4 (0.2-41)	1 (0.08-50)	<.0001
INR	1.2 (0.5-8.4)	1.3 (0.8-6.9)	1.2 (0.5-8.4)	<.0001
Hgb (g/dL]=)	11.3 (0.89-20)	10.4 (1.16- 18.6)	11.3 (0.89- 20)	<.0001
WBC (x 103/µL)	7.8 (1.3-96)	9.1 (2.7-37.6)	7.7 (1.3-96)	<.0001
Platelets (x 103/µL)	187 (2-800)	167 (20-460)	188 (2-800)	<.0001
BNP (pg/mL)	818 (1-7500)	1043 (46-5720)	809 (1-7500)	0.001
Hemodynamics
Heart Rate (bpm)	86 (40-180)	91 (48-155)	86 (40-180)	<.0001
MAP(mmhg)	76.67 (28-143)	73.67 (47.33-143)	76.67 (28-129.33)	<.0001
RAP (mmhg)	12 (-3-40)	16 (0- 40)	12 (-3-40)	<.0001
PASP (mmhg)	50 (3-100)	48 (4-96)	50 (3-100)	0.001
PADP (mmhg)	25 (0-60)	25 (1-60)	25 (0-60)	0.32
PCWP (mmhg)	24 (1-60)	25 (2-60)	24 (1-60)	0.14
Cardiac index (L/min/m2)	1.99 (0.4- 8.96)	1.89 (0.47-5.02)	1.99 (0.4-8.96)	0.004
RVSWI (g*m/m2)	0.48 (0.86-5.83)	0.33 (-0.16-2.42)	0.48 (-0.86-5.83)	<.0001
PA pulse pressure [PASP-PADP] (mmHg)	25 (-37-74)	20.5 (-14- 54)	25 (-37- 74)	<.0001
PAPI [PAPP/RAP]	2.1 (-5-55)	1.27 (-0.78-13)	2.13 (-5-55)	<.0001
RAP/PCWP	0.5 (-0.5-13)	0.63 (0-2.77)	0.5 (-0.5-13)	<.0001
TPG (mmhg)	8.67 (-39.3-57.3)	7 (-11-30.3)	9 (-39.3-57.3)	<.0001
PVR (WU)	2.17 (-9.7-25.9)	1.95 [-3.9-20)	2.18 (-9.9-25.9)	0.02
Echocardiogram
LVEDD [cm]	6.8 (1-10)	6.6 (2.6-10)	6.8 (1-10)	<.0001
Tricuspid Regurg (severe)	1092 (12.7%)	65 (19.9%)	1027 (12.4%)	<.0001
Mitral Regurg (severe)	5066 (58.1%)	180 (52.8%)	4886 (58.4%)	0.04
RV dysfunction (mod/severe-severe)	2746 (50.9%)	151 (67.7%)	2595 (50.2%)	<.0001
LVEF < 20%	6224 (70.5%)	246 (71.1%)	5978 (70.4%)	0.79
Interventions within 48h of LVAD
Hemodialysis or UF	239 (2.4%)	35 (9.1%)	204 (2.1%)	<.0001
Mechanical ventilation	597 (6.0%)	72 (18.7%)	525 (5.5%)	<.0001
IABP	2674 (26.8%)	142 (36.8%)	2532(26.4%)	<.0001
ECMO	219 (2.2%)	57 (14.8%)	162 (1.7%)	<.0001
HMII risk score ≥ 2	3868 (54.3%)	224 (77.0%)	3644 (53.4%)	<.0001
Concomitant procedure at time of LVAD				<.0001
TVR	1554 (15.6%)	69 (17.9%)	1485 (15.5%)
Other	2341 (23.5%)	143 (37.0%)	2198 (22.9%)
None	6079 (60.9%)	174 (45.1%)	5905 (61.6%)

Abbreviations. ACEI – angiotensin converting enzyme inhibitor; ALT – alanine aminotransferase; ARB – angiotensin receptor blocker; AST – aspartate aminotransferase; BNP – brain natriuretic peptide; BPM – beats per minute; CRT- cardiac resynchronization therapy; ECMO – extracorporeal membrane oxygenation; IABP – intraaortic balloon pump; ICD – implantable cardiac defibrillator; LVEDD – left ventricular end diastolic diameter; MAP – mean arterial pressure; MI- myocardial infarction; Mod – moderate; NYHA – New York Heart Association Classification; PADP – pulmonary artery diastolic pressure; PAP – pulmonary artery pressure; PAPI – pulmonary artery pulsatility index; PAPP – pulmonary artery pulse pressure; PASP – pulmonary artery systolic pressure; PCWP – pulmonary capillary wedge pressure; PVR- pulmonary vascular resistance; RAP – right atrial pressure; Regurg – regurgitation; RV – right ventricle; RVSWI – right ventricular stroke work index; TPG – transpulmonary gradient; TVR – tricuspid valve repair; UF – ultrafiltration; VAS – visual analogue scale; WU – woods units.

Risk factors for RVAD

The unadjusted odds ratios for RVAD utilization associated with baseline characteristics are available in Supplemental Table 1. The multivariable model (Table 2) demonstrated reasonable performance for predicting early RVAD utilization with c-statistic 0.78 (Figure 2). A calibration plot comparing predicted versus actual events is available as Supplemental Figure 2.

Table 2.

Multivariable predictive risk model for early RVAD

TERM	OR	95% CI	p-value
Intercept	0.04	(0.01 to 0.16)	<.0001
Prior CABG/valve surgery	1.70	(1.34 to 2.14)	<.0001
INTERMACS profile 1*	2.79	(2.00 to 3.89)	<.0001
INTERMACS profile 2*	1.98	(1.49 to 2.64)	<.0001
ECMO within 48h	2.71	(1.82 to 4.05)	<.0001
Hemodialysis or UF within 48h	1.67	(1.08 to 2.57)	0.02
Creatinine (per 1 mg/dL increase)	1.25	(1.07 to 1.46)	0.005
INR (per unit increase)	1.50	(1.02 to 2.22)	0.04
Total bilirubin (per 1 mg/dL increase)	1.13	(1.04 to 1.23)	0.003
WBC (per ×103/UL increase)	1.04	(1.01 to 1.07)	0.008
RA pressure (per 1 mmHg increase)	1.05	(1.03 to 1.08)	0.0001
Stroke volume × 100 (per unit increase)	0.89	(0.80 to 0.99)	0.03
PA pulse pressure (per 1 mmHg increase)	0.96	(0.94 to 0.98)	0.0002
LVEDD (per 1 cm increase)	0.80	(0.67 to 0.95)	0.01
Tricuspid Regurgitation (severe)	1.61	(1.18 to 2.19)	0.002
Other concomitant procedure	1.47	(1.14 to 1.88)	0.003
Concomitant TV repair	1.06	(0.78 to 1.45)	0.70

Abbreviations. CABG – coronary artery bypass graft surgery; ECMO – extracorporeal membrane oxygenation; INR – international normalized ratio; LVEDD – left ventricular end diastolic diameter; PA – pulmonary artery; RA – right atrial; TV – tricuspid valve; WBC – white blood cell count.

^*Referent group: Profiles 3–7.

Figure 2.

Receiver operating characteristic curve (ROC) of the logistic regression model for the end-point of right ventricular assist device (RVAD) use within 14 days of continuous flow left ventricular assist device implantation. Sensitivity (Sens), specificity (Spec), positive predictive value (ppv), and negative predictive value (npv) are provided at four points along the ROC curve.

Patient baseline characteristics according to strata of predicted risk for RVAD from the multivariable model are summarized in Table 3. The actual incidence of RVAD utilization was 16.2% in the highest risk decile compared to the predicted incidence of 15.8% (Supplemental Figure 2). Among patients with an estimated probability of RVAD use ≥ 10%, 73% were INTERMACS profile 1, 30% were on ECMO, 18% were on renal replacement therapies, and 72% had a concomitant procedure (Table 3). Median serum creatinine for this very high risk group was 1.5 mg/dL. Median LVEDD was 6.2 mm in those with a ≥ 10% risk compared to 7.2 mm in patients with risk < 1%. Severe TR was more common in the highest versus lowest risk strata (27% vs. 3%). Patients with an estimated risk ≥ 10 % had a median RAP of 18.6 mmHg and pulmonary artery pulse pressure (PAPP) of 17.1 mmHg, whereas the group with an estimated risk < 1% had a median RAP of 8 mmHg and PAPP of 28 mmHg. Each 1 mmHg increase in RAP was associated with a 5% increase in the risk of RVAD, while each 1 mmHg increase in PAPP was associated with a 4% decrease in risk (Table 2). Following adjustment for other risk factors, INTERMACS profile 1 was associated with 2.79 times greater risk than profiles 3–7 (Table 2). The HeartMate II RV Risk Score was not independently associated with RVAD use.

Table 3.

Profiles of Risk: Patient characteristics by prediction of risk

		Estimated probability of RVAD within 14 days of CF-LVAD
		<1%	1–<5%	5–<10%	≥10%
Total N		1359	6618	1304	695
INTERMACS profile	% profile 1	0%	7%	39%	73%
	% profile 2	5%	43%	55%	26%
	% profiles 3–7	95%	50%	6%	1%
Prior CABG/valve surgery	% prior surgery	9%	29%	39%	39%
TR regurgitation	% Severe	3%	12%	24%	27%
Creatinine (mg/dL)	Median	1.2	1.3	1.4	1.5
Total Bilirubin (mg/dL)	Median	0.8	1	1.5	2
INR	Median	1.1	1.2	1.4	1.4
WBC (× 103/µL)	Median	7	7.6	9.2	11.6
RAP (mmHg)	Median	8	12.3	17	18.6
PAPP (mmHg)	Median	28	25	21.1	17.1
Stroke volume (×100)	Median	5.8	4.7	4.1	3.9
LVEDD (cm)	Median	7.2	6.8	6.5	6.2
Hemodialysis or UF	% last 48 hours	0%	1%	4%	18%
ECMO	% last 48 hours	0%	0%	1%	30%
Concomitant procedures
TV repair	% TVR	8%	15%	24%	18%
Other	% other	11%	21%	34%	54%
	% none	81%	64%	42%	28%

Abbreviations. CABG – coronary artery bypass graft surgery; ECMO – extracorporeal membrane oxygenation; INR – international normalized ratio; LVEDD – left ventricular end diastolic diameter; PAPP – pulmonary artery pulse pressure; RA – right atrial; TV – tricuspid valve; WBC – white blood cell count.

Risk of Early RVAD or Death

There were 222 early deaths and 608 composite events including death or RVAD. The c-statistic of the multivariable RVAD model for the composite end-point model was 0.73. Actual and predicted incidence of death/RVAD increased proportionally across risk deciles, demonstrating that the model accurately discriminates risk of the composite event. Larger differences were found between actual versus predicted events, signifying that the composite model is less accurately calibrated as would be anticipated given that it was derived to predict risk of RVAD alone. Fourteen of the 16 variables from the RVAD model retained significance in the composite model and covariate beta coefficients remained stable in the new model with < 25% change in value for all but five variables. Collectively these data suggest that predictors of RVAD use have a similar association with the composite end-point of death or RVAD.

Early RVAD and Survival

Patients who received an early RVAD had decreased survival compared to those who received an isolated LVAD (log-rank for linear trend p ≤ 0.0001) (Figure 3). Thirty-day survival from the time of LVAD implantation was 73.5% versus 96.1%. The unadjusted hazard ratio for death associated with early RVAD was 3.12 (95% CI: 2.68–3.63; p ≤ 0.0001). Multivariable adjustment for other baseline risk factors did not dramatically affect this relationship [HR 2.76 (95% CI: 2.34–3.24; P ≤ 0.0001)]. The global test of proportional hazards for the COX PH model was significant, indicating that the hazard for death associated with RVAD use is not constant across time and appears greater in the early period following RVAD placement.

Figure 3.

Kaplan Meier survival curves for patients with (hashed line) and without (solid line) right ventricular assist device (RVAD) use within 14 days of continuous flow left ventricular assist device implantation. Table inserts provide the estimated survival and 95% confidence limits for patients with and without RVAD use at selected time points.

Discussion

The INTERMACS registry provides a rich resource for evaluating outcomes following LVAD surgery, and this study represents the largest analysis of RVAD use among a contemporary cohort of patients undergoing CF-LVAD surgery. A prior multi-center study of RHF after LVAD was conducted in a BTT population utilizing a single device and had only 30 RVAD events² compared to the 386 events available in this analysis. Our findings confirm many of the risk factors for early RHF after LVAD previously identified in smaller studies and provide a foundation for more precise estimation of risk regarding the need for RVAD support in LVAD candidates. As many early deaths following LVAD implantation are likely attributable to severe RHF, we also demonstrated that the model discriminated risk of the composite end-point of RVAD or death.

Outcomes with isolated LVADs are consistently better than with TAH or BIV MCS¹. When feasible, LVAD support alone is therefore the preferred MCS strategy. Furthermore, in some observational studies, delayed RVAD is associated with worse outcomes compared to early RVAD^{4, 5}. Thus, having greater accuracy to predict the need for RV mechanical support is of paramount clinical importance. Currently available models to predict RHF after LVAD had demonstrated at best modest performance in validation cohorts, and these studies have included very small numbers of RVAD recipients. This study provides an opportunity to examine risk factors specifically for RVAD utilization with higher fidelity around the risk associated with different clinical factors to help aid in clinical decision making.

There was clear evidence from multiple domains that patients requiring early RVAD support were more ill at the time of LVAD implant than those who did not require an RVAD. INTERMACS patient profiles were found to be independently associated with RVAD utilization. Profiles 1 (“crash and burn”) and 2 (“sliding on inotropes”) were associated with 2–3 times the risk of RVAD compared to profiles 4–7. However, only 10% of profile 1 and 4% of profile 2 patients required RVAD support, highlighting the need to incorporate other risk factors into LVAD patient selection.

Hemodynamic assessment has a central role in understanding RV performance, and several prior studies have identified different hemodynamic metrics as risk factors for RHF after LVAD^{2, 6–8}. While the relationship between cardiac and vascular function on arterial pulse pressure is complex, in advanced HFrEF, a narrow arterial pulse pressure is an established marker of poor hemodynamic status and a predictor of adverse outcomes^15–18. Extrapolating correlations from the systemic to the pulmonary circulation is limited by physiologic differences in the anatomy and function of these distinct chambers and vascular beds. Nevertheless, narrow PAPP appears to be a correlate of impaired RV function and was independently associated with RVAD utilization. Similar to other published studies, elevated RAP was a strong risk factor for RVAD implant^{6, 9, 10, 19}. Two prior studies identified the PAPI (PAPP/RAP) as a risk factor for RVAD^{19, 20}. While the component variables of the PAPI were predictors in the INTERMACS model, the index itself did not add incremental value to risk prediction. Similarly, RAP/PCWP ratio² had a significant univariate association with RVAD use, but in contrast to RAP alone, was not an independent predictor. However, given that some of these hemodynamic variables are highly correlated, it should be noted that alternative models with similar performance could likely be derived from this same dataset utilizing these alternative covariates.

Prior history of CABG or valve surgery was associated with increased risk of RVAD as it is likely that past cardiac surgery increases the risk of right ventricular injury, prolonged cardiopulmonary bypass time, and perioperative bleeding. Excessive transfusion can precipitate RHF by causing RV distension as well as increased PVR^{21, 22}. Concomitant surgery at the time of LVAD implant was also associated with RVAD use which may be related to prolonged cardiopulmonary bypass time, the need for cardioplegia, and/or myocardial ischemia which are detrimental to RV function.

As reported elsewhere^{23, 24}, severe pre-implant tricuspid regurgitation (TR) was associated with RVAD use. Concomitant TVR, however, was not associated with RVAD risk in the INTERMACS registry. This finding is similar to those from the Society for Thoracic Surgeons (STS)²⁵ and HeartWare bridge-to-transplant (ADVANCE) trial datasets²³. Among patients with severe TR undergoing implantation of a centrifugal-flow LVAD, those who received concomitant TVR were less likely to develop late RHF²³. Whether TR should be corrected at the time of LVAD implantation remains uncertain, but the INTERMACS data provide reassurance that TV repair does not appear to result in an increased need for RVAD support.

The association between lower LVEDD and RVAD use has been reported^{26, 27}. Patients with refractory HF and a non-dilated LV may be suffering either from a restrictive cardiomyopathy with RV involvement or from a predominantly a right-sided HF syndrome, both of which are less likely to improve with left-sided support alone. Patients with a non-dilated LV may also be more prone to the impact of the LVAD on mechanical ventricular interdependence²².

To improve clinician usability, most of the published risk scores for RHF have dichotomized continuous variables at ‘optimal cut-points’ to define patients at high risk. This approach may fail to capture continuum of risk associated with these variables by imposing a somewhat arbitrary binary threshold. Dichotomization of variables can create false reassurance when a predictor is below a given threshold. We present the INTERMACS RVAD model in its entirety with the associated odds ratios for each predictor to allow clinicians to appraise the magnitude of risk for RVAD use accompanied by different patient characteristics. This approach also avoids assigning excessive weight to a single factor, as complete understanding of RHF integrates information across multiple RV functional domains including echocardiographic, biochemical, and hemodynamic profiles. An integrated, multi-modality approach to RV assessment is important to guide decision making prior to LVAD implantation.

Establishing prohibitive thresholds of RVAD risk is challenging. These thresholds would likely vary by patient, dependent on factors including age and implant strategy. It may be reasonable to assume a higher level of risk if transplantation is possible, allowing RVAD support to be temporary should it be necessary. A long-term RVAD would be sub-optimal in a CF-LVAD recipient implanted as DT, in whom greater emphasis is placed on non-survival outcomes associated with long-term support including quality of life and functionality. DT LVAD surgery is increasingly viewed as a semi-elective procedure to be avoided in patients with critical cardiogenic shock with end-organ dysfunction, requiring ECMO and/or renal replacement therapies, all independent predictors of RVAD utilization. National trends have demonstrated a decline in LVAD implantation among profile 1 patients, consistent with avoiding LVAD utilization in these higher risk cohorts¹.

Limitations

Limitations of this study include its retrospective nature as well as those associated with registry data. There was substantial missing data in the registry. Right heart catheterization variables were missing for greater than 20% of observations, which was accounted for with multiple imputation, a standard statistical technique. Omission of these observations may introduce bias given non-differential patterns of missingness. Despite truncating extreme outliers, a few implausible data points remained. The absolute number of these outliers was small, and unlikely to impact study findings.

A major factor limiting the utility of RHF prediction includes the challenges associated with RV imaging and inability to incorporate quantitative RV functional assessment into risk models²⁸. Indeed, echocardiographic assessment of qualitative RV function was missing in greater than 50% of subjects and was excluded from analyses. Furthermore, the physiology of post-operative RHF is complex with several potential contributing pathophysiologic pathways. Intraoperative variables that predispose to RHF including RV ischemia, changes in RV volume due to transfusion, increases in PVR associated with cardiopulmonary bypass, acidosis or hypoxia, and the impact of LVAD output on RV loading, geometry and function, are not available for comparison²². Furthermore, the low prevalence of RVAD use limits the positive predictive value of prediction models. Finally, it remains unclear whether the variables identified in this analysis are relevant to other forms of RHF after LVAD surgery including the need for prolonged inotropic support.

Conclusions

This study is the largest multicenter experience exploring preoperative characteristics associated with RVAD use following CF-LVAD implant. Correlates of increased clinical acuity, including markers of end-organ dysfunction and profiles of hemodynamic instability, are associated with RVAD utilization. Preoperative prediction remains challenging given the complex nature of post-operative RHF and the low prevalence of RVAD use. The risk factors identified in our final model should supplement the clinical judgment of a multidisciplinary team during patient selection and preoperative planning for LVAD surgery.

Clinical Perspective.

What is new?

• Of 9,976 patients undergoing continuous-flow LVAD surgery in the INTERMACS registry, approximately 4% required a right ventricular assist device (RVAD) within two weeks of implantation.

• An additional 2% of patients suffered early death, many of which were likely secondary to severe right heart failure (RHF).

What are the clinical implications?

• Preoperative right ventricular assessment should incorporate data from the general clinical assessment, blood work, cardiac imaging, and hemodynamic evaluation. Abnormal trends across multiple parameters best highlight the high risk patient.

• Global markers of illness severity including evidence of end-organ dysfunction and the pre-operative need for ECMO or dialysis are associated with increased RVAD utilization. Hemodynamic profiles of RHF include elevated right atrial pressure, narrow pulmonary artery pulse pressure, and reduced cardiac output, while echocardiographic profiles include severe tricuspid regurgitation and a less dilated left ventricle.

• Intraoperative events can contribute to the risk of postoperative RHF. Concomitant cardiac surgical procedures during LVAD implantation are associated with RVAD use and require careful consideration during surgical planning.