Outcomes after tricuspid valve surgery concomitant with left ventricular assist device implantation in the EUROMACS registry: a propensity score matched analysis

Abstract

Open in new tabDownload slide

OBJECTIVES

Tricuspid regurgitation (TR) is common in patients receiving a left ventricular assist device (LVAD). Controversy exists as to whether concomitant tricuspid valve surgery (TVS) is beneficial in currently treated patients. Therefore, our goal was to investigate the effect of TVS concomitant with a LVAD implant.

METHODS

The European Registry for Patients with Mechanical Circulatory Support was used to identify adult patients. Matched patients with and without concomitant TVS were compared using a propensity score matching strategy.

RESULTS

In total, 3323 patients underwent LVAD implantation of which 299 (9%) had TVS. After matching, 258 patients without TVS were matched to 258 patients with TVS. In the matched population, hospital deaths, days on inotropic support, temporary right ventricular assist device implants and hospital stay were comparable, whereas stay in the intensive care unit was higher in the TVS cohort (11 vs 15 days; P = 0.026). Late deaths (P = 0.17), cumulative incidence of unexpected hospital readmission (P = 0.15) and right heart failure (P = 0.55) were comparable between patients with and without concomitant TVS. In the matched population, probability of moderate-to-severe TR immediately after surgery was lower in patients with concomitant TVS compared to patients without TVS (33% vs 70%; P = 0.001). Nevertheless, the probability of moderate-to-severe TR decreased more quickly in patients without TVS (P = 0.030), resulting in comparable probabilities of moderate-to-severe TR within 1.5 years of follow-up.

CONCLUSIONS

In matched patients, TVS concomitant with LVAD implant does not seem to be associated with better clinical outcomes. Concomitant TVS reduced TR significantly early after LVAD implant; however, differences in probability of TR disappeared during the follow-up period.

INTRODUCTION

Implantation of a left ventricular assist device (LVAD) improves survival, functional status and quality of life in patients with end-stage heart failure [1, 2]. In these patients tricuspid regurgitation (TR) is common [3], and current guidelines recommend consideration of tricuspid valve surgery (TVS) when moderate-to-severe TR is present [4]. Nevertheless, controversy exists whether concomitant TVS is associated with better outcomes, because contemporary studies are hampered by small sample sizes and are biased due to baseline differences [5]. In this study, we investigated the clinical outcomes after TVS concomitant with LVAD implantation compared to propensity score matched controls using the European Registry for Patients with Mechanical Circulatory Support (EUROMACS). Furthermore, we assessed the postoperative course of TR in patients with and without concomitant TVS.

METHODS

Study design

The EUROMACS is a registry of the European Association for Cardio-Thoracic Surgery. In this registry all relevant clinical, echocardiographic, haemodynamic and laboratory parameters of patients who require mechanical circulatory support have been collected prospectively since January 2011. Participating centres were allowed to enter data acquired before 2011 retrospectively, making this study an ambispective cohort study. Detailed descriptions of the database and the collection procedure were provided previously [6].

Patients

All patients operated on between 1995 and 2018 were identified. Patients <18 years old and with planned right ventricular (RV) or biventricular were excluded from analysis. Additionally, patients with single ventricle physiology were excluded (Supplementary Material, Fig. S1).

Study outcome

The main outcomes that were assessed were early (both 30-day and hospital deaths separately) and late deaths. A late death was defined as death after 30 days, regardless of hospital admission status. Furthermore, unplanned hospital readmission and right heart failure were assessed. Right heart failure was defined according to the INTERMACS adverse event definitions [7]. Patients were censored at heart transplant, death and when lost to follow-up. Lastly, the course of the probability of moderate-to-severe TR was evaluated in patients with and without TVS.

Missing values

Multiple imputation by chained equations using the statistical MICE package in R was used to impute missing values [8] Selected baseline variables with <55% missing values were imputed; >55% missing values was considered excessive missingness (Supplementary Material, Table S1). Nevertheless, 51 out of the 67 imputed variables (76%) had <30% missing values. An exception was made for the variable tricuspid annular plane systolic excursion (62% missing), because this variable is highly important in the setting of TVS, and it was reasonable to assume it could be imputed based on observed variables, such as the RV ejection fraction (missing mechanism: missing at random). Imputations were done based on the other baseline variables. In the case of highly correlated variables, the variable with the highest clinical value was chosen as the predictor (Supplementary Material, Table S2). Correlation was tested with Pearson R or Spearman rho, as appropriate. Five imputed datasets were generated with this method using 5 iterations each. The imputations were visually checked by strip plots and density plots, and no major deviations were noted between imputed data and complete data (e.g. tricuspid annular plane systolic excursion: Supplementary Material, Fig. S2). Analyses were done on each dataset separately and pooled according to Rubin’s rules [9]. In baseline comparisons of the matched groups, continuous data were transformed to the approximate Gaussian distribution and were pooled according to Rubin’s rules.

Statistical analyses

Continuous data are presented as mean ± standard deviation (Gaussian distribution) or median [interquartile range (IQR)] (non-Gaussian distribution). Categorical data are presented as frequencies (percentage). Comparisons among continuous variables were made with the Student’s t-test or the Mann–Whitney test, as appropriate. Continuous data outside 3 standard deviations were considered erroneous and removed (Supplementary Material, Table S3). Comparisons of categorical variables were made with the χ² test or with the Fisher’s exact test, as appropriate. Propensity score matching was used to balance baseline differences, because the main interest of this study is the treatment effect in a typical treated patient instead of a population level treatment effect [10]. The parsimonious propensity score model was developed using least absolute shrinkage and selection operator regression [11]. This machine learning analysis technique shrinks unimportant covariates to zero. The parsimonious model comprised all non-zero covariates. In total, 62 variables were offered to the least absolute shrinkage and selection operator model, which selected 15 variables (Supplementary Material, Table S4). Thereafter, 9 variables were added due to clinical significance and to achieve satisfactory balance (Supplementary Material, Table S5). The final propensity score model contained 24 variables (Supplementary Material, Tables S5 and S6). One-on-one matching without replacement was performed, and the caliper was set at 0.15. For the main outcome, a sensitivity analyses was performed with the caliper set at 0.001. Standard mean difference before and after matching was used to assess covariate balance. Late survival was calculated and visualized with the Kaplan–Meier method; both cohorts were compared with the log-rank test. Because some patients had no recorded follow-up, a sensitivity analysis was performed to test the robustness of the log-rank test under different missing mechanisms. Unplanned hospital readmission and right heart failure were considered competing risks with death, and Fine and Gray competing risk models were used to calculate cumulative incidences. Gray’s tests were used to quantify significant differences among cohorts. Generalized mixed models were used to analyse repeated echocardiograms. Further details regarding the mixed models are provided in Supplementary Material, Text S1. Follow-up completeness was calculated using the modified Clark C (C*) [12]. All analyses were done in R (R core team 2017, Vienna, Austria) with the use of statistical packages ‘glmnet’, ‘Matching’, ‘survival’, ‘cmprsk’, ‘splines’ and ‘lme4’.

RESULTS

In total, 3323 procedures were included [3024 (91%) without TVS and 299 (9%) with TVS]. In the TVS cohort, 292 (97%) patients had a tricuspid valve repair, and 7 (3%) patients had a tricuspid valve replacement (6 mechanical and 1 biological). After propensity score matching, 258 procedures without TVS surgery were matched to 258 procedures with additional TVS. Density plots of the propensity score in the unmatched and matched cohorts are presented in Fig. 1. In patients who survived 30 days and had recorded late follow-up information, the mean follow-up time was 1.7 ± 1.5 years with a completeness of 86% (C*).

Patient characteristics

Patient characteristics are presented in Table 1. In the unmatched cohort, patients who did not undergo TVS had, among others, significantly less TR, more ischaemic cardiomyopathy and better kidney and liver function. In the matched cohort, no significant differences in baseline characteristics were noted. In addition, the overall absolute standard mean difference before matching was 18.7 and after matching, it was 4.9 (Supplementary Material, Table S7).

Hospital outcome

Hospital outcomes are presented in Table 2. In the unmatched cohort, cardiopulmonary bypass time (80 vs 118 min; P < 0.001), intensive care unit (ICU) stay (10 vs 15 days; P < 0.001), hospital stay (30 vs 34; P = 0.001) and days on inotropic support (>14 days: 24.7% vs 32.4%) were longer in the patients who underwent TVS. In the matched cohorts, these variables were all comparable, except for cardiopulmonary bypass time (85 vs 116 min; P < 0.001) and ICU stay (11 vs 15 days; P = 0.026) (Table 2). Additionally, in the matched groups, the 30-day mortality risk [13.6%, 95% confidence interval (CI) 9.5–18.6 vs 10.0%, 95% CI 6.5–14.4; P = 0.27] and hospital mortality risk (20.2%, 95% CI 14.7–24.7 vs 16.5%, 95% CI 13.0–22.6; P = 0.41) were comparable between the patients with and without concomitant TVS. Sensitivity analyses with the caliper at 0.001 did not change point estimates greatly (Supplementary Material, Table S8).

Late outcome

In total, 2522 patients had recorded late follow-up and did not die within 30 days (no TVS: 2263 and TVS: 259 patients); 809 patients died during the follow-up period (Supplementary Material, Fig. S3). Kaplan–Meier survival curves are shown in Fig. 2A, B. Unmatched patients with and without concomitant TVS had comparable late survival rates (P = 0.41). Additionally, cumulative incidence plots are shown in Figs 3A and B and 4A and B. In unmatched patients, cumulative incidence of unplanned hospital readmission from any cause and cumulative incidence of right heart failure were higher in the TVS cohort (Figs 3A and 4A); P-value = 0.006 and P-value = 0.011, respectively.

In the matched cohort, 226 patients with TVS survived 30 days and had recorded late follow-up versus 204 matched controls, 128 of whom died during the follow-up period. Late survival was comparable between patients with and without TVS (P = 0.17) (Fig. 2B). Notably, the curves diverged after ∼1 year of follow-up with 2-year survival estimates of 75.6% (95% CI 69.3–82.5) in the no TVS cohort and 63.2% (95% CI 55.3–72.2) in the TVS cohort, but still with overlapping CIs. In total, 22 patients in the matched control group and 7 patients in the TVS cohort did not have recorded follow-up information. Sensitivity analyses revealed that only in the scenario in which all missing patients in the no TVS cohort survived and in which all in the TVS cohort died, the log-rank test results differed significantly (Supplementary Material, Table S9). Sensitivity analyses with the caliper set at 0.001 did not considerably change point estimates (Supplementary Material, Table S8). In the matched cohorts, cumulative incidence of unplanned hospital readmissions (P = 0.15) and right heart failure (P = 0.55) were comparable between patients with and without concomitant TVS (Figs 3B and 4B).

Evolution of tricuspid regurgitation

In total, 1219 patients had 3956 recorded echocardiograms during the follow-up period (mean: 3.2 echocardiograms, range: 1–28). Figure 5A presents the probability of moderate-to-severe TR over time in the unmatched cohorts. In the matched cohorts, 224 patients had 725 recorded echocardiograms (mean 3.2, range 1–21) that could be used in the mixed models. Immediately after LVAD implantation, patients who underwent TVS had a significantly lower probability of moderate-to-severe TR (33% vs 70%; P = 0.001) (Fig. 5B). Nevertheless, during follow-up, the probability of moderate-to-severe TR decreased more quickly in the no TVS cohort compared to the TVS cohort (P = 0.030), resulting in comparable probabilities within 1 year of follow-up.

DISCUSSION

We evaluated outcomes of concomitant TVS during LVAD implantation in the largest European LVAD registry. In the matched cohort, comparable risks and rates of mortality, days on inotropic support, cumulative incidence of unexpected readmission and right heart failure were noted. Not surprisingly, cardiopulmonary bypass time was longer in the TVS cohort. Furthermore, patients who underwent concomitant TVS stayed longer in the ICU compared to patients who did not undergo TVS. Immediately after surgery the probability of moderate-to-severe TVS was significantly lower in the TVS cohort; however, this difference disappeared during the follow-up period.

Patients undergoing TVS are significantly different from patients without concomitant TVS. Patients undergoing TVS presented as less acute patients with a longer history of cardiac diagnosis and fewer ischaemic aetiologies (among others), which is also illustrated by different densities in propensity scores (Fig. 1). Hence, patients undergoing TVS seemed to be a select subgroup in the overall LVAD population. It has to be noted that conclusions regarding treatment effect in this study only apply to this subgroup and may not apply in other subgroups within the LVAD population.

Prior analyses of the Society of Thoracic Surgeons database and the INTERMACS database revealed results comparable to our results [13, 14]. Patients receiving TVS who were recorded in the Society of Thoracic Surgeons database stayed longer in the ICU. RV assist device implant and hospital mortality risks were comparable in this cohort [13].

The investigators of the INTERMACS database noted comparable rates of late survival in patients with preoperative moderate-to-severe TR with and without concomitant TVS [14]. Moreover, a recent systematic review, pooling mostly small retrospective studies, found no differences in early and late survival risks/rates [5]. Interestingly, both in retrospective studies and INTERMACS database studies, it was noted that pre-LVAD moderate-to-severe TR was associated with poorer late survival rates [3, 14, 15]. Regarding the latter observation, it seems peculiar that eliminating TR does not result in a better outcome. Two hypotheses may explain these paradoxical results. First, TVS may not sustainably reduce post-LVAD TR. Song et al. found a relatively high rate of recurrent TR in patients who received concomitant TR. Additionally, there are reports that LVAD support exacerbates TR due to a leftward shift of the interventricular septum and increased venous return [16, 17]. Nevertheless, our results support the idea that TVS reduces TR soon after the operation, but that in patients without concomitant TVS, TR also decreases in the following months. Second, it may be that TR does not cause death in most cases. It is known that TR is frequently caused by RV dilatation in response to elevated pulmonary pressures [18]. Therefore, TR may merely be a symptom or a marker of RV damage secondary to long-standing pulmonary hypertension or primary RV damage caused by the underlying ischaemic or cardiomyopathic diseases. By treating TR, one may be treating the symptom rather than the causing factor of mortality and morbidity (e.g. RV dysfunction). To some extent, our findings support this theory because favourable RV remodelling is observed in patients with an LVAD implant without concomitant TVS [19, 20]. This finding would inherently be paired with a reduction of TR, even without an intervention, assuming that the TR is functional in nature.

In this respect, the cause of TR (primary or secondary) is important. Primary TR, caused by structural valve damage or interfering pacemaker/implantable cardioverter defibrillator leads, will certainly not reduce itself and may even cause RV dysfunction [21]. Therefore, we propose that this aspect be taken into account in the decision process whether to perform concomitant TVS. Robertson et al. [13] suggested that the decision to perform concomitant TVS should not be solely based on the pre-LVAD TR grade. Our data and reports in the current literature support this suggestion, because our results and multiple other studies were do not to show any benefit from concomitant TVS. Current guidelines suggest consideration of concomitant TVS in all patients with pre-LVAD moderate-to-severe TR, which may not be necessary. Nevertheless, if one follows the trends in concomitant TVS for functional TR during left-sided valve surgery, it has become clear that TR in some cases is not reduced or even becomes worse [22, 23]. The remaining challenge is now to adequately identify these patients in the LVAD population.

Strengths and limitations

The strength of this study is the relatively large sample size compared to those reported in the current literature. Additionally, the EUROMACS registry records serial echocardiograms, which made it possible to analyse the change in TR over time. In contrast to previous studies, we accounted for the within-patient correlations in our analyses of the postoperative course of TR over time using advanced statistical modelling. This study has several important limitations. First, the database is not designed to specifically address concomitant TVS in patients with LVAD implants. Therefore, important factors, such as the cause of TR or the reasons for intervention, were not collected. This lack may introduce selection bias, because these factors could not be captured in the propensity model. Furthermore, the surgeon and institutional preferences can introduce selection bias. Although the majority of variables of interest had below 30% missing values, we accepted up to 55% missing values. On the other side, the EUROMACS database collects many variables, making it more plausible that missing data could be predicted from the other observed variables, thereby strengthening the missing-at-random assumption. In addition, since last year, the EUROMACS investigators intensified their quality control measures to reduce missingness in the future [24]. Furthermore, assessing TR remains challenging: TR is subject to loading conditions, which means TR severity is highly dynamic [25]. Unfortunately, it was impossible to analyse patients receiving a tricuspid valve replacement compared to a tricuspid valve repair due to the small numbers. Some differences could be due to chance because of multiple testing. Propensity score matching reduces the sample size and therefore may reduce the power of the tests. Nevertheless, we utilized a matching technique because the main interest of this study was the effect of treatment in a typical treated patient. Some patients in the matched population had no recorded follow-up information. Nevertheless, sensitivity analyses did not change the direction of the conclusions in most of the hypothetical missing scenarios.

CONCLUSIONS

Patients undergoing concomitant TVS differ significantly from patients without TVS. In matched patients, concomitant TVS during LVAD implant does not seem be associated with better clinical outcomes. Concomitant TVS reduced TR significantly early after LVAD implant; however, differences in probability of TR disappeared during the follow-up period. Using current selection criteria, TVS does not seem beneficial.

Conflict of interest: none declared.

REFERENCES

Author notes