Surgery for prosthetic valve endocarditis: a retrospective study of a national registry^†

Abstract

INTRODUCTION

Prosthetic valve endocarditis (PVE) is a serious complication of heart valve replacement, reported to occur with an incidence of 0.3–1.2% per patient-year, corresponding to 3–6% of all patients receiving a prosthetic valve within 5 years of implantation [1–3]. Despite the advancements in medical perioperative and surgical care, PVE still carries postoperative mortality rates that are among the highest in cardiothoracic surgery [4, 5], along with poor long-term outcomes in hospital survivors and high health system costs [5, 6].

Although representing a relevant proportion of all cases of infective endocarditis (IE), the relatively low incidence of PVE in the general population have been causing management guidelines to be based on scarce amount and quality of evidence [1]. In particular, PVE is recognized to be a totally different clinical entity compared with native valve endocarditis [4, 5]; however, today in many aspects, the treatment of PVE simply follows the same principles as for native valve endocarditis. To overcome the low numbers and provide robust analysis of outcomes and their determinants, multicentre series are necessary.

The present study aimed at describing the clinical profile (i.e. preoperative features) and postoperative course of PVE in a large patient population and at identifying the independent predictors of early outcome of surgery, exploiting a national registry established by the Research Group on Outcomes in Cardiac Surgery (GIROC; Supplementary Appendices A and B) of the Italian Society for Cardiac Surgery (SICCH).

MATERIALS AND METHODS

Patient population

The present study took advantage of a national multicentre database of surgical IE, which currently includes 3718 consecutive patients operated on for IE of either a native or a prosthetic heart valve in 26 Italian cardiac surgery centres (Supplementary Appendix A) between 1979 and 2015. For 7 centres, the completion of the database is still ongoing, with only minimal required data being already inserted; therefore, for the purpose of the present study, the initial population consisted of the 2823 patients from the first 19 enrolling centres. Of them, only the patients with PVE, in whom retrospectively the modified Duke criteria were satisfied [1, 7], were considered: the definitive study cohort comprised 582 patients. Redo patients who had received valve replacement in their history but developed endocarditis of a native valve thereafter were not considered in the present series.

Study variables

The database includes 102 principal variables and 39 derived variables. Those included in the analysis regarded: patient’s demographics, era of operation (categorized into 3 groups: 1980–2000; 2001–08; and 2009–14), microbial data (positive/negative/unavailable blood/valve cultures and class of pathogen isolated), activity phase (defined according to Duke’s criteria [1, 7]), patient’s comorbidities (chronic obstructive pulmonary disease, diabetes, cirrhosis, dialysis and previous pacemaker implantation) and risk factors (as included in the EuroSCORE), additive and logistic EuroSCORE, type of infected prosthesis (homograft/autograft, bioprosthesis or mechanical prosthesis), echocardiographic data (abscess, vegetation, peri-prosthetic leak, cusp tear and ejection fraction), operative data (type of prosthesis implanted, number of valves replaced, cardiopulmonary bypass time and cross-clamp time, associated procedures), 30-day postoperative complications (re-opening for bleeding, sepsis, low cardiac output, multi-organ failure, renal failure/substitution, atrioventricular block, atrial fibrillation, malignant arrhythmias, intra-aortic balloon pump, extracorporeal membrane oxygenation, permanent and temporary neurologic dysfunction and respiratory failure) and hospital (30-day) death. The comorbidity variables in the data set were defined according to the EuroSCORE criteria, wherever applicable.

Statistical analysis

The outcome variables were in-hospital (30-day) postoperative mortality and morbidity. The morbidity outcome was defined as the endpoint of any one or more of the postoperative complications listed above. All continuous variables were tested for normality of distribution and presented as mean ± standard deviation if normally distributed, median and 25th–75th percentiles (interquartile range) if asymmetrically distributed. Categorical variables were summarized as counts and percentages. Univariate analyses were performed to assess differences among the 3 study eras (1979–2000, 2001–08 and 2009–14) by analysis of variance with Bonferroni posthoc test, χ² or Kruskal–Wallis test and associations of baseline and perioperative factors with the endpoints (by unpaired t-test, χ² or Mann–Whitney test). All variables significantly associated to the endpoints in univariate analysis (P < 0.05) and variables that were clinically relevant and with a P-value of <0.1 were introduced in multivariable logistic regression models as covariates, and the forward stepwise selection method was used. The fit of the final regression models was verified by means of the Hosmer and Lemeshow test. Accuracy was tested by calculation of the area under the receiver operating characteristic curve.

RESULTS

Clinical profile

The preoperative clinical profile of the PVE patient population is summarized in Table 1. Information on previously implanted prosthetic types was available only for 62% patients. Data on the causative pathogen was present only for 53% because blood or tissue samples had not been collected or sent for microbiological examination in 47% overall. In 59 patients (10%), both a previously implanted prosthesis and a native valve were involved by the infection. For 70 patients (12%), PVE surgery was a re-redo procedure, while for other 4 (0.7%) a third redo.

The distribution of the study patients over the years is shown in Fig. 1. When the 3 study eras were compared, a worsening preoperative clinical profile was observed in the more recent periods, with significant differences in terms of patient age (56 ± 14, 64 ± 15, 65 ± 14 years respectively, P < 0.001), preoperative embolic stroke (6%, 5%, 11%, P = 0.045) and comorbidities (median logistic EuroSCORE: 14% [7–31], 21% [8–52], 23% [9–47], P = 0.025). Also, a decreasing prevalence of female sex (47%, 42%, 34%, P = 0.012), a tendency to slight decrease in prevalence of Staphylococcusaureus (19%, 18%, 15%, P = 0.23) and to slight increase in enterococcal aetiology (10%, 15%, 17%, P = 0.21) were observed.

Figure 1

Chronological distribution of the 582 procedures for prosthetic valve endocarditis (PVE) included in this study.

Open in new tabDownload slide

Operative data

In 486 patients (83.5%), 1 valve was replaced, in 85 (14.6%) 2 valves (either 2 prostheses or 1 prosthesis and 1 native valve) were treated, in 11 (1.9%) 3 valves. Operative data are summarized in Table 2. Across the 3 chronological eras, an increasing frequency of double or triple valve surgery was observed (11%, 16%, 19%, P = 0.049).

Early outcomes: univariable analysis

Hospital (30-day postoperative) mortality was 19.2% (112 patients). Causes of death were reported as ‘cardiac cause’ in 48 patients, ‘sepsis’ or ‘multi-organ failure’ in 22 patients, ‘others’ in 16 (including neurological complications in 6, respiratory failure in 4, uncontrollable bleeding in 3 and acute renal failure in 3). In 26 cases, the information on causes of death was not available. Across 3 eras (1980–2000, 2001–08 and 2009–14), early mortality did not significantly change (20.4%; 17.1%; 20.5%, respectively, P = 0.60). The following variables showed a significant association with mortality in univariable analysis: older age (65 ± 13 years vs 62 ± 15 in hospital survivors, P = 0.045), female sex (mortality 23% vs 13% in male patients, P = 0.03), renal insufficiency defined as preoperative serum creatinine >2 mg/dl (37% vs 18%, P = 0.004), chronic obstructive pulmonary disease (35% vs 18%, P = 0.009), preoperative shock (48% vs 15%, P < 0.001), preoperative intubation (44% vs 18%, P = 0.002), mitral valve replacement (24% vs 16%, P = 0.02), tricuspid valve repair/replacement (38% vs 18%, P = 0.02), triple valve surgery (54% vs 26% double valve, 17% single valve, P = 0.004), while streptococcal aetiology was associated with lower mortality (11% vs 21%, P = 0.03), as well as healed endocarditis (13% vs 21%, P = 0.012). The median logistic EuroSCORE was 16% [6–39%] in survivors, 40% [18–64%] in patients who died in the hospital (P < 0.001). Variables showing a weaker association with mortality included: hypertension (24% vs 18%, P = 0.075), heart failure (26% vs 18%, P = 0.08), abscess (26% vs 18%, P = 0.09) and diabetes (26% vs 19%, P = 0.17). There was no significant difference in mortality among different types of prosthesis implanted to replace infected prostheses (aortic: 21% for mechanical, 15% for biological, 14% for homografts, P = 0.29). Staphylococcal aetiology did not imply significant disadvantage in terms of mortality (20% vs 15%, P = 0.26).

Complications occurred postoperatively in 256 patients (44%), including 26 patients who died in hospital and whose records were lacking data about type of complications and cause of death. The most frequent complications were acute kidney injury in 48 patients (8.2% overall; 19% of patients experiencing complications), requiring replacement therapy in 22 cases, and low output syndrome in 45 patients (7.7%), followed by re-exploration for bleeding (27 patients, 4.6%), respiratory failure (24 patients, 4.1%), sepsis (20 patients, 3.4%), stroke (17 patients, 2.9%), need for pacemaker implantation (12 patients, 2.1%), temporary neurological deficit (10 patients, 1.7%), multi-organ failure (8 patients, 1.4%), malignant ventricular arrhythmias (5 patients, 0.9%) and others (including: atrial fibrillation, pneumonia, pneumothorax, intestinal bleeding, peripheral embolism, pulmonary embolism, post-cardiotomy syndrome, reoperation for early peri-prosthetic leak and cerebral haemorrhage). Of note, 43% of patients experiencing complications of any type died within 30 days. The increase in the observed mortality with complications such as stroke and low cardiac output syndrome was nearly twice the one implied by other complications such as acute postoperative respiratory failure or kidney injury (Fig. 2).

Figure 2

Mortality rates in prosthetic valve endocarditis (PVE) patients experiencing postoperative complications. The black bold line represents the mortality in the overall series (19.2%). Note that for 26 patients the information about the complication preceding hospital death was lacking. AKI: acute kidney injury; LOS: low output syndrome; Resp Fail: respiratory failure; PMK: bradiarrhythmias requiring permanent pacemaker implantation; TND: temporary neurological deficit; Malign Arrh: ventricular fibrillation or asystolia.

Open in new tabDownload slide

The composite outcome of postoperative morbidity was significantly associated in univariable analysis with: study era (1980–2000: 18.5%, 2001–08: 38.2%, 2009–14: 52.8%, P < 0.001), increasing patient age (65 ± 14 vs 61 ± 15 years, P = 0.001), chronic pulmonary disease (56% vs 36%, P = 0.036), Enterococcus spp. (65% vs 41%, P = 0.002), lower ejection fraction (49.8 ± 10 vs 52.3 ± 10, P = 0.004) and preoperative shock (63% vs 41%, P = 0.001). Healed endocarditis was associated with significantly lower rate of postoperative complications (29% vs 48%, P < 0.001), whereas female sex (P = 0.08) and preoperative renal failure (P = 0.09) failed to reach a significant association.

Predictors of early outcomes in multivariable analysis

In multivariable logistic regression analysis, mortality was independently associated with increasing patient age, preoperative renal insufficiency, preoperative shock and triple valve surgery, whereas streptococcal endocarditis, higher ejection fraction and healed endocarditis were associated with survival (Table 3). The same results were obtained when hypertension, diabetes and abscess were included among covariates. We did not include the logistic EuroSCORE variable in the final model to avoid multi-collinearity issues. A good fit was demonstrated by the Hosmer–Lemeshow test (χ² = 4.7, P = 0.79), and the area under the receiver operating characteristic curve was 0.73 ± 0.03 (Fig. 3). The multivariable logistic regression model predicting postoperative complications was implemented including the era of operation as a covariate, to adjust for the difference observed in univariable analysis for this outcome. Patient factors that were independently associated in multivariable analysis with postoperative complications were female sex, Enterococcus spp. and preoperative shock, whereas higher ejection fraction was associated with uncomplicated postoperative course (Table 4). The Hosmer–Lemeshow test showed a χ² of 5.9 (P = 0.66) and the area under the receiver operating characteristic curve was 0.70 ± 0.02.

Figure 3

Receiver operating characteristic (ROC) curve of the regression model predicting mortality in 582 surgically treated prosthetic valve endocarditis. AUC: area under the curve.

Open in new tabDownload slide

DISCUSSION

Of the few recent outcome studies including cohorts of PVE patients, some consisted of mixed surgically and medically treated patients groups [1, 3, 8], others included also native valve endocarditis patients [9, 10]. To the best of our knowledge, the present study features the largest series of exclusively surgical PVE patients ever reported in literature. Although descriptive in nature, this study sets a benchmark for comparisons with reported series of medically treated PVE, considering the lack and scarce feasibility of randomized trials. Moreover, it can provide guidance for future studies, indicating whether and where current criteria and standards of management can be improved.

This study confirms that surgery for PVE is a high-risk one; however, the overall 30-day mortality rate was 19.2%, which is in the lower range of the reported mortality rates from other smaller PVE series. Of note, 51% patients had a logistic EuroSCORE  ≥20%. In 2010, a Chinese registry included 80 consecutive PVE patients treated over a 7-year period: mortality in the 34 surgically treated patients was 15% [8]. An earlier series from two centres reported 17% mortality with surgical therapy in a total of 104 PVE patients [11]. Among single-centre studies, Romano et al. reported 24% mortality in a series of 95 PVE patients (60% active endocarditis) [4] and Musci et al. in 349 PVE patients, all operated on in active phase, observed an hospital mortality of 28% [5]. In the International Collaboration on Endocarditis Prospective Cohort Study (ICE-PCS), including 490 surgically treated patients, the unadjusted mortality was 22% at 30 days [2].

The clinical profile of patients with PVE undergoing surgery in our country, as depicted in our present analysis, has to be considered reflective of a milieu including different referral patterns. In tertiary centres, patients were sent for operation either just after evaluation or after a variable time period of medical treatment; in other institutions, an earlier referral allowed surgeons to be systematically involved earlier in joint patient evaluation and management planning. Another recent large study exclusively including surgically treated patients [5] had a more homogeneously tertiary care-like referral pattern than ours: likely related to the inherent delayed referral for operation, preoperative profiles were worse and hospital mortality was higher (28.4%), but the study did not report on postoperative complications.

The fact that the present study spanned over 35 years allowed us to compare consecutive periods, thus investigating the time-related trends in presentation and outcomes. At odds with some previous reports, the proportion of PVE operations with streptococcal aetiology in our series increased, though not significantly, from the first (25%) to the second era (33%), to remain stable in the third (32%). In this perspective, the rate of missing microbiological data decreased significantly over the 3 eras (79%, 45% and 39%), likely reflecting not only a better possibility to collect recent than old data but also an actual improvement in preoperative definition of the aetiology. A proportion of our missing microbiological data, however, is surely related to cases in which, due to tertiary referral for emergent/urgent surgery, establishing a preoperative microbiological diagnosis had been impossible. Indeed, although the rate of Staphylococcusaureus PVE in this series was similar as in the surgical arm of the ICE-PCS [2], it was lower than in previous reports [5, 6]. Besides possible geographical reasons, another explanation may lie in the known association of Staphylococcusaureus with the worse clinical presentations of PVE, with higher incidence of emergency operation and consequent unavailability of pathogen definition: in other terms, a proportion of staphylococcal PVE cases may be hidden in the group with missing microbiology data.

The patient population with PVE, consistent with the epidemiological trends of IE in general [12, 13], grew progressively older over the chronological eras, especially from the first to the second one. As shown by the logistic EuroSCORE, preoperative conditions were progressively worse through the 3 eras. However, 30-day mortality rates were stable (20%, 17% and 20%), despite increasing EuroSCORE and age and also despite the increasing rates of postoperative complications (18%, 38% and 53%). This suggests that not only surgery but also postoperative care has improved over time. As an example, the mortality in patients with postoperative kidney injury decreased from 42% to 25% from the second (2000–08) to the third era (2009–14), and in patients with postoperative sepsis from 80% to 41%; less significant improvements however were observed for complications like low postoperative output syndrome, malignant arrhythmias and postoperative stroke.

The strongest predictors of mortality in multivariable logistic regression were triple valve surgery and preoperative shock, demonstrating that the haemodynamic conditions are the main determinants of early outcome for PVE surgery, which is further emphasized by ejection fraction being a ‘protective’ factor. Since the cardiopulmonary bypass time and cross-clamp time failed to show association with unfavourable outcome, triple valve surgery may have played as a surrogate indicator of worse cardiac and circulatory conditions, rather than being an index of more complex and time-consuming surgery. The importance of preoperative haemodynamic status to PVE surgery outcomes is confirmed by heart failure [2, 8, 11], mechanical support and high preoperative catecholamine doses [5] being independent predictors of mortality in previous series. A proportion of preoperative shock cases were indeed related to preoperative septic state: the association with staphylococcal aetiology nearly reached significance (P = 0.06).

The association of Streptococcus PVE with lower mortality, also confirmed in multivariable analysis, is consistent with an earlier report from the multicentre ICE-PCS study [14]. Conversely, we found no association between Staphylococcusaureus and mortality, whereas studies mixing surgically and medically treated PVE [8, 11, 15] consistently observed an unfavourable prognostic role of the staphylococcal aetiology. More aggressive surgical indications are recommended by the latest guidelines for Staphylococcusaureus PVE [1]. The lack of association of Staphylococcus with mortality in the present surgical registry, as well as in the Berlin series [5], may corroborate this recommendation [1, 15]. Consistent with this, a bad prognostic role of paravalvular involvement (typically abscess), as found in mixed medical and surgical series [3], was not confirmed in the present analysis. The relatively low prevalence of abscess in the present series must be acknowledged, with consequent lesser need for aortic root replacements than in other series [5] and is probably explained by a more often timely referral of patients for surgery.

Whether a type of valve substitute is superior to another in IE treatment is an open debate [5, 16]: since in the present Italian registry the follow-up data collection is still underway, and there are many missing data on prosthetic type implanted, we cannot answer the question by now. Strikingly, for the aortic position, early mortality in mechanical valve recipients was non-significantly higher than in bioprosthesis or homograft patients, notwithstanding their significantly younger mean age (57 ± 14, 69 ± 12, 60 ± 17 years; P < 0.001). When stratifying for single versus multiple valve replacement, however, the difference was confirmed only for double/triple valve replacement patients (mortality in simple aortic valve replacement being 15% for mechanical, 14% for biological and 13% for homografts). Greater numbers and more importantly the information on late recurrence of PVE will be necessary to perform an adequately adjusted analysis.

To our knowledge, this is the first study to focus on incidence, type, prognosis and determinants of complications after surgery for PVE. However, we had to cumulate all types of events together, because of the low numbers for each of the most important and frequent complications, preventing an individual analysis of the respective risk factors. Again preoperative shock and ejection fraction were significantly independently associated with the outcome. This indicates that when antibiotic therapy fails to prevent infection spreading with sepsis or when cardiac function is seriously affected by valve dysfunction and/or infection itself, PVE has reached a stage that severely threatens patient’s prognosis even after surgical treatment. Postoperative low cardiac output, sepsis and stroke were the complications with the highest mortality rates: of them, low output syndrome should be most dreaded inasmuch as the second most frequent complication after kidney injury. The other factor associated with postoperative complications in logistic regression analysis was Enterococcus spp., which to our knowledge is an unprecedented result: the prevalence of Enterococcus spp. in the present series was indeed slightly higher than in other reports [10, 17], consistent with a similar observation in another Italian study, focusing on native valve endocarditis [13]. Probably our high numbers allowed the Enterococcus variable to emerge as a multivariable predictor of complications, inasmuch as associated with significantly older patient age (70 ± 9 years vs 62 ± 15, P < 0.001), obesity (12% vs 3%, P = 0.01), chronic pulmonary disease (16% vs 6%, P = 0.03) and lower ejection fraction (47 ± 10 vs 52 ± 10, P = 0.001). The types of complication significantly associated with Enterococcus PVE were acute kidney injury (18% vs 7%, P = 0.013) and temporary neurologic deficit (8% vs 1%, P = 0.006), suggesting an association with frailty and metabolic disorders.

The present study has a number of inherent limitations: these surely include, like in all multicentre studies, heterogeneity in defining variables, in regional referral and practice patterns, and in data collection methods. Since this was not a population-based study, we cannot draw conclusions on the absolute trends in PVE incidence in Italy. For a similar reason, the study design did not allow us to address the question whether a valve type is more or less prone to develop PVE. Moreover, the odds ratios for microbiological variables cannot be relied upon strictly, because of the relatively high proportions of missing data about the pathogen. Some variables had too many missing data and had to be removed from the analysis, including time from first valve operation to PVE, time between PVE diagnosis and operation, drug abuse, antibiotic sensitivity of the pathogens. As an exemplary consequence, we could not distinguish between early and late PVE. However, as highlighted by the ESC guidelines [1] the distinction is sometimes artificious as the important difference between the two entities is in patient conditions and bacterial agents rather than in time since previous operation. Antibiotics are an important contribution in the management of surgically treated PVE: failure to be able to include modalities and types of antibiotic regimens in the analysis was another limitation. Finally, the follow-up is still to be completed for many of the involved centres, therefore we limited the analysis to in-hospital outcomes: we expect that once completed with long-term data, the database will be a useful source of data for further investigations.

In conclusion, this large multicentre study confirmed that PVE is a serious condition but also supported the current official recommendations to promptly treat surgically the complicated forms of PVE. Notwithstanding the worsening preoperative patient profile over time, current surgical methods and standards of perioperative care can contain mortality, even in patients developing complications. Preservation of the haemodynamic status in general and cardiac function in particular should be the main goal in anticipation of surgical referral, as these represent the main factors influencing outcomes. Further effort is warranted to understand the clinical and prognostic differences between PVE forms caused by the different microbial agents, so to tailor the therapeutic approach accordingly.

SUPPLEMENTARY MATERIAL

Supplementary material is available at EJCTS online.

Conflict of interest: none declared.

REFERENCES

Author notes