Abstract
Continuous-flow left ventricular assist devices (CF-LVADs) are central to advanced heart failure management but are complicated by right ventricular failure (RVF) in up to 40% of patients, increasing morbidity and mortality and underscoring the need for robust prediction tools. We performed a single-center retrospective study of CF-LVAD recipients from March 2009 to May 2024. Of 326 patients, 205 met inclusion criteria; RVF was defined as need for inotropes or right ventricular assist device support. We compared established RVF risk scores (Michigan, Penn/Fitzpatrick, European Registry for Patients with Mechanical Circulatory Support [EUROMACS], Central venous pressure, severe Right ventricular dysfunction, preoperative Intubation, severe Tricuspid regurgitation, Tachycardia [CRITT]) using receiver operating characteristic analysis, logistic regression, and internal validation. Among 205 patients, 81 (39.5%) developed post-LVAD RVF. The EUROMACS score had the highest predictive value (C-statistic 0.670, P < .001), followed by CRITT (0.653, P < .001), Penn (0.616, P = .004), and Michigan (0.606, P = .005). Youden-optimized cutoffs were derived to summarize sensitivity, specificity, positive predictive value, and negative predictive value. Mortality analysis was performed as a secondary exploratory endpoint. Overall, discrimination of existing RVF scores remained modest. EUROMACS performed best but highlighted the need to refine models with contemporary hemodynamic and echocardiographic metrics and to pursue rigorous external validation.
Introduction
Continuous flow left ventricular assist devices (CF-LVADs) have revolutionized the management of advanced heart failure, offering life-saving options for patients as destination therapy (DT), a bridge to transplant (BTT), or a bridge to orthotopic heart transplant (OHT) candidacy.1,2 Despite their transformative potential, LVADs come with significant challenges, particularly the occurrence of intraoperative bleeding and right ventricular failure (RVF) in the early postoperative period. RVF affects up to 40% of patients and necessitates additional interventions, including the use of right ventricular assist devices (RVADs) in ~5% of cases. This complication is strongly associated with higher mortality, morbidity, and healthcare costs.3 -6
Predicting and mitigating RVF are essential to optimizing patient outcomes. Risk stratification allows clinicians to determine the feasibility of LVAD implantation and to plan for potential RVAD support. Factors influencing RVF include patient-specific clinical variables, preoperative optimization, and the surgical technique used during LVAD implantation. Prediction models, such as the Michigan, 7 Penn (Fitzpatrick), 8 European Registry for Patients with Mechanical Circulatory Support (EUROMACS), 9 and Central venous pressure, severe Right ventricular dysfunction, preoperative Intubation, severe Tricuspid regurgitation, Tachycardia (CRITT) scores 10 incorporate echocardiographic and invasive hemodynamic data to assess RVF risk. The characteristics of these scores are summarized in Table 1. However, many of these models have undergone limited external validation and may not adequately capture the complexity of RVF predictors.11,12
Characteristics of Some Existing Models for Post-LVAD RVF.
Abbreviations: AST, aspartate aminotransferase; BiVAD, biventricular assist device; CI, cardiac index; CRITT, Central venous pressure, severe Right ventricular dysfunction, preoperative Intubation, severe Tricuspid regurgitation, Tachycardia; CVP, central venous pressure; d, days; EUROMACS, European Registry for Patients with Mechanical Circulatory Support; Hb, hemoglobin; HR, heart rate; INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LVAD, left ventricular assist device; MCS, mechanical circulatory support; NO, nitric oxide; PCWP, pulmonary capillary wedge pressure; RAP, right atrial pressure; RV, right ventricle/right ventricular; RVAD, right ventricular assist device; RVF, right ventricular failure; RVSWI, right ventricular stroke work index; SBP, systolic blood pressure; TR, tricuspid regurgitation.
This primary objective of the present study was to evaluate and compare discrimination of 4 established RVF risk scores (Michigan, EUROMACS, Penn/Fitzpatrick, and CRITT) 12 for predicting post-LVAD RVF in a contemporary single-center cohort of CF-LVAD patients, highlighting their clinical applicability and limitations. Secondary objectives include examining discrimination of selected preoperative clinical/hemodynamic predictors and mechanical circulatory support, assess score performance after accounting for clinical severity and temporal factors, and perform internal validation. Mortality analysis was conducted as secondary exploratory evaluations to assess whether RVF risk stratification correlates with death outcomes in this cohort.
Methods
Study Design
This single-center retrospective cohort study analyzed CF-LVAD patients treated at the University of Iowa Health Care between March 31, 2009 and May 1, 2024. The institutional review board approved the study. Patient data were extracted from electronic medical records using the Slicer Dicer tool (Epic Systems Corporation, Verona, WI, USA).
Study Population and Outcome Definition
The cohort included adult patients (≥18 years) who underwent CF-LVAD implantation. Of the 326 patients identified, 205 met the inclusion criteria: age ≥18 years and undergoing CF-LVAD for the first time. Exclusion criteria included prior LVAD implantation or incomplete clinical data (Figure 1). Data collected encompassed baseline cardiac risk factors, cardiac history, baseline testing, perioperative interventions, and transfusion needs. The primary outcome was post-LVAD RVF, defined using a composite endpoint of inotropes use ≥14 days post-LVAD implantation, and/or postoperative RVAD requirement. For sensitive analysis, RVF was further categorized by timing (early ≤14, 15-30, and late >30 days) when timing data were available.
Flow diagram of included patients.
Mortality was evaluated as a secondary exploratory outcome. Due to the unavailability of last known alive/last follow-up dates for the survivors in the cohort extract, formal time-to-event survival analysis and censoring estimation were not performed. Mortality was summarized descriptively and using fixed post-implant windows (30, 90 days, and 1 year), as well as time-to-death among decedents.
Statistical Analysis
Baseline characteristics were summarized by RVF status. Categorical variables were analyzed using the chi-square test, while continuous variables were assessed with the Wilcoxon rank-sum test. We evaluated discrimination of the 4 RVF risk scores (Michigan, EUROMACS, Penn/Fitzpatrick, and CRITT) 12 using receiver operating characteristic (ROC) curved and area under the curve (AUC, C-statistic) with 95% confidence intervals. The Michigan RVF Risk Score uses 4 preoperative variables (vasopressor requirement for 4 points, AST ≥80 IU/L for 2 points, total bilirubin ≥2.0 mg/dL for 2.5 points, and creatinine ≥2.3 mg/dL or renal replacement therapy for 3 points) with a maximum of 11.5 points, stratifying patients into low (≤3.0), intermediate (4.0-5.0), and high (≥5.5) risk categories. The EUROMACS score is a 9.5-point scale incorporating INTERMACS class 1 to 3 (2 points), multiple intravenous inotropes (2.5 points), severe RV dysfunction on echocardiography (2 points), RA/PCWP ratio >0.54 (2 points), and hemoglobin ≤10 g/dL (1 point), with scores >4 points indicating high risk. The Penn/Fitzpatrick score incorporates 6 variables including RV stroke work index, cardiac index, severe RV dysfunction, elevated creatinine, previous cardiac surgery, and systolic blood pressure ≤96 mmHg. The CRITT score is an acronym for 5 binary variables: CVP >15 mmHg, severe right ventricular dysfunction, preoperative intubation, severe tricuspid regurgitation, and tachycardia with heart rate >100/min. In addition to unadjusted ROC analysis, logistic regression models were used to estimate effect sizes (β-coefficients and odds ratio [OR] with 95% confidence intervals) for RVF. To address potential temporal confounding across the 15-year inclusion period, implant year was included as an adjusted covariate and era-based sensitivity analysis was performed (pre-HeartMate (HM) 3 vs HM3 era). Models were additionally evaluated with INTERMACS severity adjustment (INTERMACs 1-3 vs 4-5).
Model calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test and Brier score. Youden-optimized cutoffs were derived for each score, and sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and false-positive rate was reported at these thresholds. Internal validation was performed using bootstrap optimism correction (B = 500) to report optimism-corrected C-statistics. Clinical utility was assessed using decision curve analysis by comparing net benefit of the best-performing model against treat-all and treat-none strategies across clinically relevant threshold probabilities.
To assess potential overlap between individual hemodynamic predictors and score components, multicollinearity was evaluated using variance inflation factors (VIF), with VIF values >5 considered suggestive of problematic collinearity. We additionally conducted a sensitivity analysis restricting individualized predictor models to a non-overlapping predictor set excluding variables embedded within published risk scores to reduce the risk of circularity in predictor-outcome evaluation. All analysis were conducted using R version 4.5.2 (R Foundation for Statistical Computing, Vienna, Austria).
Results
Patient Characteristics and Outcomes
Of the 326 patients initially identified, 121 were excluded for reasons such as being younger than 18 years, previous LVAD implantation, or missing data (Figure 1). The remaining 205 patients included 81 (39.5%) who developed RVF post-LVAD.
Baseline Characteristics
Baseline characteristics and clinical outcomes are detailed in Table 2. Patients who developed RVF (n = 81) differed from those without RVF (n = 124) primarily by device era and severity, and preoperative hemodynamics. Device type varied significantly. HMII was less common in the RVF group than in the no RVF group (41% vs 60%; P = .006), while HM3 was more common in the RVF group than in the no RVF group (52% vs 33%; P = .007). RVF cases had greater clinical acuity, with a higher proportion of INTERMACS 2 (43% vs 21%; P < .001) and a lower proportion of INTERMACS 5 (3.7% vs 12%; P = .036).
Baseline Characteristics of the Included Patients. Comparison of RVF Versus No RVF.
Abbreviations: AFIB/AFLUT, atrial fibrillation/atrial flutter; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; CABG, coronary artery bypass grafting; CAD, coronary artery disease; CIT, cardiac index by thermodilution; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; COT, cardiac output by thermodilution; DOB, dobutamine; DPG, diastolic pulmonary gradient; HM3, HeartMate 3; HMII, HeartMate II; HR, heart rate; IABP, intra-aortic balloon pump; ICD, implantable cardioverter-defibrillator; INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LVAD, left ventricular assist device; LVDD, left ventricular end-diastolic diameter; LVEF, left ventricular ejection fraction; MCS, mechanical circulatory support; MI, myocardial infarction; MIL, milrinone; PAD, pulmonary artery diastolic pressure; PAM, pulmonary artery mean pressure; PAPi, pulmonary artery pulsatility index; PAS, pulmonary artery systolic pressure; PCWP, pulmonary capillary wedge pressure; PTCA, percutaneous transluminal coronary angioplasty; RA, right atrial pressure; RVAD, right ventricular assist device; RVF, right ventricular failure; SD, standard deviation; VFIB/VTACH, ventricular fibrillation/ventricular tachycardia; WBC, white blood cell count.
Mean ± SD; n (%).
Wilcoxon rank sum test; Pearson’s chi-squared test.
At admission, RVF cases had worse right-sided filling pressures and pulmonary pressures, including higher right atrial (RA) pressure (16.4 ± 5.7 vs 12.9 ± 5.8 mmHG; P < .001), pulmonary artery systolic (PAS) pressure (58.6 ± 14.8 vs 51.8 ± 15.5 mmHg; P < .001), pulmonary artery diastolic (PAD) pressure (29.9 ± 7.2 vs 27.2 ± 9.0 mmHg; P = .006), mean pulmonary artery (PA) pressure (41.0 ± 9.0 vs 36.1 ± 11.1 mmHg; P < .001), and pulmonary capillary wedge pressure (PCWP; 28.5 ± 6.3 vs 25.8 ± 9.3 mmHg; P = .009). Transpulmonary gradient (TPG) was also higher in RVF cases (12.0 ± 7.5 vs 9.0 ± 8.2 mmHg; P = .007), while pulmonary artery pulsatility index (PAPi) showed a nonsignificant trend toward lower values (2.0 ± 1.2 vs 2.3 ± 1.5; P = .08). After hemodynamic optimization, between-group differences were attenuated and no longer statistically significant for most optimized measures (all P > .1).
Use of pre-LVAD support was more frequent among RVF cases, including any mechanical circulatory support (MCS; 56% vs 31%; P < .001), intra-aortic ballon pump (IABP; 49% vs 26%; P < .001), Impella (9.9% vs 3.2%; P = .047), and inotrope use (75% vs 57%; P = .008). Laboratory differences included higher bilirubin (1.2 ± 0.8 vs 2.3 ± 15.0 mg/dL; P = .008) and lower hemoglobin (11.0 ± 2.1 vs 11.7 ± 2.1 g/dL; P = .017) among RVF cases. Postoperative right ventricular assist device (RVAD) utilization and chest left open were substantially more common in the RVF group (41% vs 0% and 33% vs 14%, respectively; both P < .001).
Performance of Predictor Scores for Post LVAD-RVF
The discriminatory performance of the 4 established RVF risk scores is summarized in Table 3, with ROC curves shown in Figure 2. Overall discrimination was modest across models. The EUROMACS preoperative risk score demonstrated the highest unadjusted discrimination (AUC 0.670, 95% CI 0.595-0.745, P < .001; n = 202), followed by the CRITT model (AUC 0.653, 95% CI 0.579-0.727, P < .001; n = 205), the Penn (Fitzpatrick) model (AUC 0.616, 95% CI 0.538-0.693, P = .004; n = 204), and the Michigan/RVF Risk Score (AUC 0.606, 95% CI 0.533-0.679, P = .005; n = 205). As shown in Table 3, the analytic sample size (n) varied slightly by score due to missingness.
Area Under the ROC Curve for Prediction of Post-LVAD RVF.
AUC CIs were estimated using the nonparametric DeLong method. P values test the null hypothesis that AUC = 0.5. n varies by score due to missingness.
Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval; CRITT, Central venous pressure, severe Right ventricular dysfunction, preoperative Intubation, severe Tricuspid regurgitation, Tachycardia; EUROMACS, European Registry for Patients with Mechanical Circulatory Support; n, number of patients; RVF, right ventricular failure.
Individualized ROC curves for the 4 scores for post LVAD-RVF.
Adjusted Models, Calibration, and Internal Validation
To account for potential temporal confounding across the study period, each RVF risk score was evaluated in an implant-year adjusted logistic regression model (score + implant year) in Supplementary Table 4. In these models, the score term remained significantly associated with RVF for all 4 scores, with the strongest association observed for CRITT (OR 1.99, 95% CI 1.44-2.81, P < .001) and EUROMACS (OR 1.47, 95% CI 1.25-1.76, P < .001), followed by Michigan (OR 1.17, 95% CI 1.04-1.32, P = .011) and Penn (OR 1.03, 95% CI 1.01-1.04, P = .0018). Implant year was independently associated with RVF in the EUROMACS (OR 1.12, 95% CI 1.03-1.23, P = .013), Penn (OR 1.09, 95% CI 1.00-1.18, P = .046), and CRITT (OR 1.10, 95% CI 1.02-1.20, P = .022) models, with a similar trend for Michigan (OR 1.07, 95% CI 0.99-1.16, P = .088). Discrimination in implant-year adjusted models remained modest, with adjusted AUCs of 0.713 (EUROMACS), 0.683 (CRITT), 0.655 (Penn), and 0.644 (Michigan).
Given heterogeneity in clinical acuity, models were additionally evaluated with INTERMACS severity adjustment (INTERMACS 1-3 vs 4-5). After adjustment, the score term remained significant for EUROMACS (OR 1.42, 95% CI 1.18-1.74, P < .001), Penn (OR 1.02, 95% CI 1.01-1.04, P = .0063), and CRITT (OR 1.81, 95% CI 1.28-2.60, P = .001), while the Michigan score was attenuated and no longer statistically significant (OR 1.10, 95% CI 0.96-1.26, P = .191). INTERMACS 1 to 3 status was independently associated with RVF in the Michigan (OR 3.00, 95% CI 1.22-8.03, P = .021), Penn (OR 3.20, 95% CI 1.42-8.03, P = .0077), and CRITT (OR 2.48, 95% CI 1.06-6.41, P = .046) models.
Model calibration was assessed using the Hosmer-Lemeshow (HL) test and Brier score (Supplementary Table 6). Calibration was acceptable for Michigan (HL P = .708), Penn (HL P = .854), and CRITT (HL P = .628). The EUROMACS model showed evidence of miscalibration (HL P = .024), although Brier scores were similar across models (range 0.203-0.221).
Internal validation using bootstrap optimism correction (B = 500) demonstrated minimal optimism across models (mean optimism ~0.006-0.010) in Supplementary Table 7. Optimism-corrected C-statistics were 0.705 for EUROMACS, 0.677 for CRITT, 0.645 for Penn, and 0.635 for Michigan, indicating that discrimination remained modest after accounting for optimism.
Cutoffs and Predictive Values (Youden Thresholds)
To improve bedside interpretability beyond AUC, Youden-optimized cutoffs were derived for each score and corresponding operating characteristics were calculated (Supplementary Table 2). The Michigan/RVF Risk Score cutoff of 3.5 yielded high sensitivity (0.773) but low specificity (0.423), corresponding to a false-positive rate of 0.577, PPV 0.436, and NPV 0.764. The EUROMACS cutoff of 4.25 showed a more balanced profile (sensitivity 0.680, specificity 0.646; PPV 0.531; NPV 0.774). The Penn (Fitzpatrick) cutoff of 49 produced higher specificity (0.708) but lower sensitivity (0.473), with PPV 0.479 and NPV 0.702. The CRITT cutoff of 1.5 demonstrated moderate sensitivity (0.613) and specificity (0.677), with PPV 0.523 and NPV 0.752. Overall, these threshold-based metrics highlight the trade-off between sensitivity and false-positive burden, particularly for the Michigan score, and motivated additional evaluation of clinical utility using decision curve analysis.
Decision Curve Analysis
Decision curve analysis was performed for the best-performing adjusted model (EUROMACS + implant year) to evaluate clinical utility across threshold probabilities (0.05-0.50) relative to treat-all and treat-none strategies (Figure 3). Across a broad range of clinically relevant thresholds, the EUROMACS + implant-year model demonstrated positive net benefit compared with treat-none, and it provided greater net benefit than treat-all over much of the lower-to-mid threshold range, supporting potential clinical usefulness despite modest discrimination and imperfect specificity.
Decision curve analysis.
Individualized Parameters
We evaluated the discriminatory performance of selected preoperative clinical and hemodynamic parameters as individual predictors of post-LVAD RVF, including MCS and invasive hemodynamics at admission (Supplementary Table 1). Among admission hemodynamics, right-sided pressures demonstrated modest discrimination, with RA pressure showing the strongest performance (AUC 0.669, 95% CI 0.594-0.744, P < .001), followed by PAS (AUC 0.642, P < .001), mean pulmonary artery pressure (PAM; AUC 0.641, P < .001), PAD (AUC 0.615, P = .004), PCWP (AUC 0.609, P = .006), and TPG (AUC 0.612, P = .005). In contrast, PAPi demonstrated limited discrimination (AUC 0.427, P = .082), and DPG was not discriminatory (AUC 0.508, P = .841).
These findings suggest that admission hemodynamic burden is more informative for post-LVAD RVF risk, while established composite risk scores maintain modest discrimination across the cohort (Table 3).
Mortality Prediction
Mortality was also evaluated as a secondary exploratory endpoint. In the overall cohort (N = 205), 80 deaths (39.0%) were observed. Cumulative mortality was 2.0% at 30 days (4/205), 6.8% at 90 days (14/205), and 14.6% at 1 year (30/205).
Discrimination of the RVF risk scores for mortality was modest and not statistically significant overall (Table 4 and Figure 4). The AUCs were 0.575 for EUROMACS (P = .066), 0.553 for CRITT (P = .172), 0.514 for Penn (P = .742), and 0.502 for Michigan (P = .961), with n varying slightly by score due to missingness.
Area Under the ROC Curve for Mortality Prediction. Secondary Exploratory Analysis.
AUC CIs were estimated using the nonparametric DeLong method. P values test the null hypothesis that AUC = 0.5. n varies by score due to missingness.
Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval; CRITT, Central venous pressure, severe Right ventricular dysfunction, preoperative Intubation, severe Tricuspid regurgitation, Tachycardia; EUROMACS, European Registry for Patients with Mechanical Circulatory Support; RVF, right ventricular failure.
Individualized ROC curves for the 4 scores for mortality.
RVF Timing and Phenotype Subgroup (Sensitivity Analysis)
Using the primary RVF definition (postoperative RVAD and/or ≥14-day inotropes), score performance was additionally evaluated across RVF timing and phenotype subgroups (Supplementary Table 3). Discrimination remained modest but varied by subgroup. For early RVF (≤14 days), discrimination was similar across scores (AUC range ~0.636-0.668, all P ≤ .013), with CRITT and Penn demonstrating the highest AUCs. For RVF occurring 15 to 30 days, CRITT and EUROMACS demonstrated modest discrimination (AUC 0.677 and 0.668, respectively; both P ≤ .001), while Michigan showed weaker performance (AUC 0.577, P = .155). For late RVF (>30 days), EUROMACS showed the highest discrimination (AUC 0.693, P = .002), whereas CRITT, Michigan, and Penn were not statistically significant.
When stratified by phenotype, RVAD-RVF demonstrated modest discrimination across all scores (AUC ~0.663-0.687, all P ≤ .003). In contrast, for prolonged inotrope-RVF, EUROMACS (AUC 0.662, P < .001) and CRITT (AUC 0.643, P = .001) showed modest discrimination, while Michigan and Penn were weaker and not statistically significant. Overall, these subgroup analysis support variability in score performance by RVF timing and phenotype and justify the primary composite RVF endpoint for the main analysis.
Discussion
Identifying patients at high risk for RVF is important since its occurrence is associated with longer hospitalization, increased cost, and worse patient outcomes.13,14 Effective prediction of RVF risk can therefore influence the choice of suitable treatment, minimize resource use, and enhance a patient’s quality of life. Most predictor scores, aside from the EUROMACS score, were generally assessed in limited single-center studies using diverse definitions of RVF, and using various LVAD models.4,9 This resulted in inconsistent predictors and a lack of a reliable model for RVF prediction. Additionally, there is a significant shortage of external validation studies for these risk prediction models. 15 In this study, we performed a comparative analysis between Michigan, Penn, EUROMACS and CRITT scoring, and found that the EUROMACS risk score performed best in predicting RVF. To address temporal confounding across the long inclusion window, we incorporated implant year in adjusted models, and score associations generally persisted. We also accounted for clinical acuity using INTERMACS severity adjustment, which influenced effect estimates for some scores. Internal validation using bootstrap optimism correction demonstrated minimal optimism, suggesting that observed discrimination is unlikely to be explained by overfitting. The EUROMACS score incorporates various factors, such as cardiopulmonary hemodynamic and echocardiographic data, patient history, and preoperative medical care. 9 Yet, our data showed only modest discrimination, indicating the presence of additional variables that can potentially account for RVF risk. Furthermore, exploratory mortality analysis did not demonstrate strong discrimination.
Calibration assessment showed acceptable fit for most models, although EUROMACS demonstrated evidence of miscalibration on the Hosmer-Lemeshow test, highlighting that discrimination alone does not ensure accurate risk estimation. Threshold-based performance (Youden cutoffs) and decision curve analysis further illustrated trade-offs between sensitivity and specificity and demonstrated net benefit for the best-performing adjusted model across clinically relevant threshold probabilities. A systematic review and meta-analysis of observational studies of risk factors associated with RVF after LVAD implant showed that patients on ventilatory support or continuous renal replacement therapy (CRRT) are at high risk for post-LVAD RVF, as well as patients with slightly increased international normalized ratio (INR), high N-terminal pro-B-type natriuretic peptide (NT-proBNP), or leukocytosis. 16 Analysis of some individualized parameters in our study suggested limited discrimination for MCS alone, while admission hemodynamics showed modest discriminatory ability.
Careful patient selection is crucial to mitigate the risk of RVF in individuals undergoing LVAD implantation. Those identified with a high score may necessitate perioperative optimization of RV support through strategies aimed at reducing preload,17,18 afterload, 19 or may require early intervention with RV mechanical support or a biventricular assist.20,21 A single-center retrospective analysis of 128 patients who underwent first CF-LVAD implantation revealed that the inability to achieve hemodynamic optimization goals was associated with higher rates of RVF and increased 1-year all-cause mortality post-LVAD implantation. 22 Despite thorough risk assessment and medical management, some patients inevitably progress to RVF, necessitating RVAD support, which is linked to worse outcomes. However, elective RVAD procedures are associated with improved long-term survival compared with emergency interventions, 23 and they enhance survival rates to transplant when compared with delayed RVAD insertion. 24 Therefore, the application of a valid RVF prediction score allows better preparation for high-risk patients and a more swift response to potential complications, while long term RV support strategies are actively being developed. 25
Developing risk scores to accurately predict RVF remains a challenge. In addition to the challenges in assessing RV function, there are factors which are difficult to integrate in a predictive scoring model and which play a major role in RVF occurrence. These include intraoperative complications, such as bleeding, and post-operative changes in pulmonary hemodynamics and device settings. 12 The lack of a universally accepted definition and classification of RVF is an additional complicating factor, so reaching a consensus definition can be a step forward.
Main Physiopathologic Mechanisms in Risk Scores Account for Right Ventricular Failure
The prognostic scores for post LVAD RVF incorporate variables that reflect the fundamental pathophysiologic mechanisms driving post-LVAD RV decompensation. These mechanisms center on hemodynamic overload, ventricular interdependence, end-organ dysfunction, and baseline RV contractile reserve.9,26,27 The EUROMACS score emphasizes markers of illness severity (INTERMACS class), preload mismatch (RA/PCWP ratio), intrinsic RV dysfunction (severe RV dysfunction on echocardiography), inotropic dependence (multiple inotropes), and oxygen-carrying capacity (hemoglobin), reflecting the RV’s inability to handle the acute increase in venous return and preload that occurs when the LVAD augments systemic cardiac output.9,26 The Michigan score prioritizes hemodynamic instability and end-organ hypoperfusion, capturing the consequences of reduced RV stroke work and elevated central venous pressure, which signal inadequate RV-pulmonary artery coupling and systemic congestion. 15 The Penn and CRITT scores similarly integrate hemodynamic parameters (central venous pressure [CVP], pulmonary artery pressures), echocardiographic indices of RV function (tricuspid annular plane systolic excursion [TAPSE], fractional area change), and biochemical markers of end-organ injury (liver enzymes, creatinine), all of which reflect the multifactorial insults to the RV including leftward septal shift (which removes septal contribution to RV contractility), increased RV wall stress, pulmonary hypertension, and perioperative stressors such as ischemia, vasoplegia, and volume resuscitation.26 -29 Collectively, these scores attempt to quantify the preoperative RV reserve and the likelihood that the RV will fail to adapt to the geometric and hemodynamic changes imposed by LVAD support, though their modest discriminatory performance underscores the complex, multidimensional nature of RV failure pathophysiology that cannot be fully captured by preoperative variables alone.
Study Limitations
The present study has several important limitations. First, it was conducted as a retrospective analysis at a single tertiary care center, which may limit the generalizability of our findings. Second, the relatively modest sample size restricted statistical power, particularly for subgroup analysis, and may have contributed to the wide confidence intervals observed for several predictors. Third, although we adhered to established definitions of RVF used in prior scoring systems, these may not fully capture late-onset RVF or the evolving spectrum of postoperative right ventricular dysfunction. Fourth, device heterogeneity over the study period (transitioning from HMII and HeartWare to HM3) introduces potential confounding, as RVF incidence and perioperative management strategies have changed over time. Fifth, incomplete hemodynamic or echocardiographic data for some patients may have introduced bias and limited the evaluation of certain individualized predictors. Sixth, calibration was imperfect for some models, which limits the accuracy of absolute risk estimation even when discrimination is modest. Seventh, multiple comparisons were performed across scores and subgroups, and results should be interpreted as exploratory. Finally, as with all retrospective studies, unmeasured confounding variables—such as intraoperative factors, postoperative device management strategies, or evolving surgical practices—could not be fully accounted for, potentially influencing the observed associations. In addition, lack of last-known-alive/last-follow-up dates precluded formal time-to-event survival modeling for mortality, and mortality analyses should be interpreted as secondary exploratory.
Conclusion
The EUROMACS score showed the highest predictive value among evaluated models for RVF post-LVAD. However, substantial gaps remain in predictive accuracy, underscoring the need for effective predictive tools. Combining traditional risk scores with novel hemodynamic and echocardiographic metrics represents a promising avenue for future research. Rigorous external validation is essential to establish the clinical utility of these models.
Supplemental Material
sj-docx-1-ang-10.1177_00033197261453467 – Supplemental material for Performance of Risk Scores in Predicting Right Ventricular Failure After LVAD Implantation
Supplemental material, sj-docx-1-ang-10.1177_00033197261453467 for Performance of Risk Scores in Predicting Right Ventricular Failure After LVAD Implantation by Ibrahim Mortada, Mohammed Mhanna, Shiva Raju Gollapally Krishna, Abdul Qadeer, Ahmad Al-Abdouh, Ernesto Ruiz Duque, Shareef Mansour and Hani Jneid in Angiology
Footnotes
Author Contributions
All authors contributed to: (1) substantial contributions to conception and design, or acquisition of data, or analysis and interpretation of data, (2) drafting the article or revising it critically for important intellectual content, and (3) final approval of the version to be published.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Supplemental Material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
