This is a retrospective analysis of prospectively collected data from an observational endocarditis cohort ENVALVE (ENdovascular and cardiac VALVE infection Cohort) in a tertiary center (University Hospital Zurich—USZ) in Switzerland. Data collection forms containing demographic, clinical, laboratory, and treatment information are completed by physicians and study nurses since January 2018. Decision on operability or rejection is made by the USZ Endocarditis Team according to the most actual Clinical Practice Guidelines and is based on patients' characteristics at the time of presentation (Modified Duke criteria, age, clinical condition, neurological status, likelihood of survival with surgery, quality of life) (1, 12) and patients' wish.

ENVALVE participants aged 18 or older with ALSIE of a native or a prosthetic valve and with confirmed indication for surgery according to current Clinical Practice Guidelines (1) and at least two cohort visits from 1 January 2018 through 31 December 2021 were included. Patients were stratified in two groups for further comparison: group 1) operative indication and subsequent surgery, and group 2) operative indication but no surgery performed. We excluded patients with isolated right-sided IE, intracardiac device-related IE and aortic graft infection from the analysis.

The modified Duke Criteria and the proposed modification from the European Society of Cardiology in 2015 were used for the diagnosis of IE since it has been shown that these criteria have a high sensitivity (1, 14). Timing for surgery was divided in three different scenarios: elective (defined as surgery after 1–2 weeks of antimicrobial therapy); urgent (within a few days, <7 days); or emergent (within 24 h) irrespective of antimicrobial treatment duration (1). Operative mortality was defined according to the Society of Thoracic Surgeons (15). Charlson comorbidity index (CCI) (16), EuroSCORE II (17) and the APORTEI predicted mortality risk score (18) were calculated to estimate perioperative morbidity and mortality. The latter has been lately validated as a similar risk stratification score for IE (18). Perioperative complications were reported as per guidelines for reporting mortality and morbidity after cardiac valve interventions (19).

Infection with hepatitis-B-Virus (HBV) was confirmed in patients with positive hepatitis-surface antigen and positive HBV DNA in laboratory testing, or in those under medical treatment. Infection with hepatitis C-Virus (HCV) was confirmed in patients with positive serological testing and detectable HCV RNA or in those under medical treatment. The Child-Pugh Score was calculated for the preoperative assessment of cirrhosis mortality.

Statistical analyses were conducted using Stata 17.0. Qualitative variables were expressed as numbers and percentages. Quantitative variables were expressed as median and interquartile range (IQR). For qualitative variables Fisher's exact test and for continuous variables the Mann-Whitney-U test were used, whenever applicable. Statistical significance was achieved when p < 0.05. Kaplan-Meier estimates were calculated for survival and are depicted with 95% confidence intervals. Log rank pairwise comparisons were performed to determine different distributions.

We assessed factors associated with survival including surgery using a bivariable Cox model. Many risk factors influence both the decision to undergo surgery and survival. To adjust for this confounding by indication we performed a multivariable logistic regression to summarize the probability of receiving surgery in a propensity score including age, sex, CCI, diabetes mellitus, hepatitis B, liver cirrhosis, cerebrovascular disease, dialysis, estimated glomerular filtration rate (eGFR) at diagnosis, creatinine at diagnosis, malignancy, ischemic stroke, haemorrhagic stroke, Staphylococcus aureus, Streptococcus spp, intubation, cardiogenic shock, septic shock, EuroSCORE II and APORTEI score. The score was then included as an interaction term in the Cox models for surgery, used both as untransformed linear variable as well as after stratification into quintiles (20).

The Ethics Committee/Institutional Review Board approved the study within the framework of the ENVALVE-cohort (BASEC 2017-01140). Written informed consent was obtained from all participants.

Between January 2018 and December 2021, we identified 146 patients with definitive ALSIE and 8 patients with possible ALSIE and a confirmed operative indication. Out of them, 70% had indication to surgery because of hemodynamic deterioration, 45% because uncontrolled infection and 25% to prevention embolization (more than one indication possible). Overall, 116 patients were operated (group 1), and 38 were denied for surgery and received medical treatment only (group 2).

The timing of surgery of the 116 patients who were operated was elective (72%), emergent (22%) and urgent (6%), and the median interval from diagnosis to surgery was 16 days [interquartile range (IQR) 8.5–28.5]. Table 1 shows the prevalence of native-valve endocarditis (NVE) and prosthetic valve endocarditis (PVE) as well as baseline characteristics. Differences between the two groups were found in age [62 years (IQR 48–70) vs. 70 years], diabetes mellitus (13 vs. 29%), eGFR (75 ml/min (50–97) vs. 41 ml/min (33–84)), dialysis (1% vs. 9%), positive hepatitis B virus (HBV) status (2 vs. 13%), and end-stage liver disease. Furthermore, differences were observed in CCI (2 units (1–5) vs. 6 units (3–9), EuroSCORE II (6% (3–21) vs. 13% (7–29)) and APORTEI mortality scores (12% (8–25.7) vs. 33% (22–52). There were also differences in hemorrhagic stroke (8 vs. 21%), renovisceral embolism (24 vs. 42%), intubation (9 vs. 26%), cardiogenic shock (11 vs. 26%) and septic shock (9 vs. 29%).

Reasons to deny surgery are shown in Table 2. There was more than one reason for surgical denial in 13 (34%) patients. The most common cause was an expected high perioperative mortality (30% of patients).

Median follow-up was 283 days (IQR 32–455). Figure 1 depicts survival estimates at follow-up. At 1 year, 84% (CI 95% 77–91) in group 1 and 17% (CI 95%, 3–30) in group 2 were alive (log-rank test <0.0001).

Univariable analyses showed that surgery had a very strong positive effect on death prevention (HR 0.11, 95% CI 0.062–0.202, p < 0.001). We also examined the effect of surgery in bivariable Cox models and all risk factors without finding relevant modifications (Figure 2). The point estimates for the hazard ratios based on the propensity score and the bivariable analysis with different additional variables varied between 0.11 and 0.18 for the endpoint surgery.

Uni- and multivariable odds ratios (OR) for surgery are shown in Figure 3. Univariable logistic regression showed associations for surgery with streptococcal IE [OR 3.03 (95% CI 1.2–7.8)] and eGFR at diagnosis [OR 1.3 (95% CI 1.1–1.5)]. There was also a trend for a positive association with ischemic stroke [OR 1.98 (95% CI 0.88–4.45)]. There was evidence of negative association of surgery with age [OR 0.57 (95% CI 0.41–0.79)], diabetes mellitus [OR 0.36 (95% CI 0.15–0.88)], hepatitis B [OR 0.18 (95% CI 0.04–0.77)], liver cirrhosis [OR 0.06 (95% CI 0.01–0.28)], cerebrovascular disease [OR 0.35 (95% CI 0.13–0.98)], dialysis [OR 0.03 (95% CI 0.003–0.27)], malignancy [OR 0.42 (95% CI 0.18–0.99)], CCI [OR 0.35 (95% CI 0.14–0.89)] and S. aureus [OR 0.36 (95% CI 0.17–0.76)]. Surgery was less likely among patients with high surgical risk scores (EuroSCORE II [OR 0.78 (95% CI 0.65–0.97)]; APORTEI [OR 0.63 (95% CI 0.51–0.79)], and also among patients with complications due to IE such as hemorrhagic stroke [OR 0.32 (95% CI 0.11–0.89)], necessity for intubation and ventilation [OR 0.29 (95% CI 0.11–0.76)], cardiogenic shock [OR 0.35 (95% CI 0.14–0.89)] and septic shock [OR 0.32 (95% CI 0.07–0.39)].

In the final multivariable adjusted model, surgery was less likely among patients with liver cirrhosis [OR = 0.03 (95% CI 0.00–0.30)] and with dialysis [OR = 0.014 (95% CI 0.00–0.47)]. There was also trend to less frequent surgery in patients with ischemic (OR 3.24 [95% CI 0.86–12] and hemorrhagic stroke (OR 0.19 [95% CI 0.04–1.0]. Because of correlations between several of these variables, it is difficult to judge the contribution of individual components in a multivariable model adjusted for all variables. However, this multivariable model allowed for a very good prediction of individual propensity scores to receive surgery (ROC AUC = 0.92).

We examined the positive effect of surgery on survival in a Cox model adjusted for the probability of surgery with the propensity score (in quintiles). Surgery continued to show a strong positive effect overall [HR 0.15 (95% CI 0.07–0.35)] and adjusted for quintiles (HR 0.20 [95% CI 0.09–0.43). Participants in the most favourable quintile had a median probability of surgery of 99% with a high rate of survival after one year (94%), whereas surgical candidates in the least favourable quintile had a median probability of surgery of 10% with a one-year mortality of 73% (Figure 4).

Strengths of our study are the prospective collection of all IE cases, risk factors and outcomes and the use of structured event reporting forms. This is a single-center, non-randomized, observational study with a limited number and heterogeneous group of patients. Further limitations are the imbalances both group (116 who underwent and 38 who received medical treatment only. This may have introduced bias. By assessing additional factors in bivariable models, none of these parameters seemed to introduce an unexpected shift in the point-estimates for the treatment effect of surgery. As this is the experience of a tertiary-care referral center, the population patterns and data may not be generalizable. There is no formal record linkage with other hospitals. Therefore, it cannot be completely excluded that clinical events (e.g., recurrent surgery, death) were not reported by participants. Outcome data are limited to one-year; therefore, we cannot make any statement about the long-term prognosis of patients.