Author + information
- Received May 18, 2013
- Revision received July 4, 2013
- Accepted July 12, 2013
- Published online October 8, 2013.
- Sandrine Hubert, MD∗,†,
- Franck Thuny, MD, PhD∗,‡,§∗ (, )
- Noemie Resseguier, MD‖,
- Roch Giorgi, MD, PhD‖,
- Christophe Tribouilloy, MD, PhD¶,#,
- Yvan Le Dolley, MD∗,
- Jean-Paul Casalta, MD∗∗,
- Alberto Riberi, MD†,
- Florent Chevalier, MD¶,
- Dan Rusinaru, MD¶,
- Dorothée Malaquin, MD¶,
- Jean Paul Remadi, MD††,
- Ammar Ben Ammar, MD‡‡,
- Jean Francois Avierinos, MD∗,
- Frederic Collart, MD†,
- Didier Raoult, MD, PhD‡∗∗ and
- Gilbert Habib, MD∗∗∗ ()
- ∗Département de Cardiologie Hôpital Universitaire de la Timone, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Marseille, France
- †Service de Chirurgie Cardiaque, Hôpital Universitaire de la Timone, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Marseille, France
- ‡URMITE, UM63, CNRS 7278, IRD 198, Inserm 1095, Aix-Marseille Université, Marseille, France
- §Département de Cardiologie, Hôpital Universitaire Nord, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Marseille, France
- ‖LERTIM and UMR 912 INSERM-IRD, Faculté de Médecine, Aix-Marseille Université, Marseille, France
- ¶Département de Cardiologie, Hôpital Universitaire Sud, Amiens, France
- #INSERM, ERI 12, Amiens, France
- ∗∗Laboratoire de Microbiologie, Centre Hospitalier Universitaire, Hôpital de la Timone, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Marseille, France
- ††Département de Chirurgie Cardiaque, Hôpital Universitaire Sud, Amiens, France
- ‡‡Département d'Anesthésie, Hôpital Universitaire Sud, Amiens, France
- ↵∗Reprint requests and correspondence:
Prof. Franck Thuny, Département de Cardiologie, Hôpital de la Timone, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Boulevard Jean Moulin, 13005 Marseille, France.
- ↵∗∗Prof. Gilbert Habib, Département de Cardiologie, Hôpital de la Timone, Assistance Publique Hôpitaux de Marseille, Aix-Marseille Université, Boulevard Jean Moulin, 13005 Marseille, France.
Objectives The aim of this study was to develop and validate a simple calculator to quantify the embolic risk (ER) at admission of patients with infective endocarditis.
Background Early valve surgery reduces the incidence of embolism in high-risk patients with endocarditis, but the quantification of ER remains challenging.
Methods From 1,022 consecutive patients presenting with definite diagnoses of infective endocarditis in a multicenter observational cohort study, 847 were randomized into derivation (n = 565) and validation (n = 282) samples. Clinical, microbiological, and echocardiographic data were collected at admission. The primary endpoint was symptomatic embolism that occurred during the 6-month period after the initiation of treatment. The prediction model was developed and validated accounting for competing risks.
Results The 6-month incidence of embolism was similar in the development and validation samples (8.5% in the 2 samples). Six variables were associated with ER and were used to create the calculator: age, diabetes, atrial fibrillation, embolism before antibiotics, vegetation length, and Staphylococcus aureus infection. There was an excellent correlation between the predicted and observed ER in both the development and validation samples. The C-statistics for the development and validation samples were 0.72 and 0.65, respectively. Finally, a significantly higher cumulative incidence of embolic events was observed in patients with high predicted ER in both the development (p < 0.0001) and validation (p < 0.05) samples.
Conclusions The risk for embolism during infective endocarditis can be quantified at admission using a simple and accurate calculator. It might be useful for facilitating therapeutic decisions.
Embolic events are frequent and life-threatening complications occurring in more than 50% of patients with infective endocarditis (1–5). They are believed to be caused by fragmentation of vegetations, but they are also dependent on the prothrombogenic conditions associated with or related to the infection (6,7). These events are factors of poor prognosis, especially with involvement of the cerebral circulation bed (2). Two types of embolic events must be differentiated: 1) embolic events occurring before the initiation of antibiotic therapy, which could be prevented by earlier diagnosis and rapid initiation of antibiotics (8); and 2) embolic events occurring during and after antibiotic therapy, which could be prevented by valve surgery (9,10). Thus, the evaluation of the embolic risk (ER) at admission is crucial in the management of endocarditis to avoid such potential catastrophic events. Several factors have been associated with ER, such as the length and localization of vegetations, the causative microorganism, and the presence of previous emboli (11,12). Thus, the current international guidelines recommend early valve surgery according to these predictors (13,14). However, these recommendations are based on a low level of evidence, and the rate of embolic events remains high despite their routine use (10,15). Recently, a randomized trial demonstrated that early surgery significantly reduced the risk for systemic embolism (10). However, this study included only patients with very low operative risk, which may limit its application in clinical practice, in which the benefit/risk ratio for surgery must be evaluated. Thus, an appropriate and accurate quantification of ER associated with a correct evaluation of operative risk would allow a more reliable assessment of the benefit/risk ratio of valve surgery to prevent embolism and help standardize endocarditis management.
Currently, there is no simple method to accurately quantify ER. Moreover, the effects of competing events, such as valve surgery or death, have always been ignored in previous studies aimed at evaluating ER. Indeed, valve operation or death, regardless of the cause, may preclude the occurrence of embolic events.
Therefore, the objective of this multicenter study was to develop and validate a model to quantify ER at admission, accounting for competing risks.
This study took place in the departments of cardiology of 2 university-affiliated tertiary care hospitals for adults in France (Marseille and Amiens), which are referral centers for their regions on infective endocarditis. At these centers, more than half of the patients are referred for complicated endocarditis with a potential indication for surgery. A specific database was created to prospectively collect information on patients with a diagnoses of suspected infective endocarditis. To ensure that all endocarditis episodes were enrolled, suspected patients were screened weekly by cardiologists and microbiologists. For this study, we reviewed all consecutive patients with definite diagnoses of infective endocarditis determined at each center, according to the modified Duke criteria (16), from January 2000 to June 2011. The exclusion criteria were isolated pacemaker or defibrillator lead endocarditis and patients already included in the study but rehospitalized for recurrent episodes of endocarditis. Thus, this study relied on incident cases of endocarditis. Written informed consent was obtained from all participating patients under an approved protocol, as required by the institutional review board.
The following data were collected at admission and during hospital stay: age, sex, Charlson comorbidity index (17), diabetes, history of cancer, intravenous drug use, underlying heart disease, chronic renal insufficiency, aspirin and/or anticoagulant therapy, causative pathogen (determined by blood cultures, serology testing, valve culture, or polymerase chain reaction on a valve specimen according to international guidelines ), heart failure, and indications for valve surgery. History of paroxysmal, persistent, or permanent atrial fibrillation was also collected at admission according to international guidelines (18). Transthoracic and transesophageal echocardiographic studies were performed in all patients within 24 h of admission, and the data from the first echocardiographic study were collected as previously described (12). Briefly, echocardiographic data included the presence and maximal length of vegetations. Vegetation length was measured in various planes, and the maximal length was used. For multiple vegetations, the largest length was used for analysis. Periannular complications were defined as an abscess, pseudoaneurysm, or fistula, according to accepted definitions (14). The severity of valvular regurgitations was assessed according to international guidelines (19).
Embolic events occurring before the initiation of antibiotic therapy (previous embolism) were systematically screened at admission by clinical examination. Moreover, cerebral, thoracic, and abdominal computed tomographic scans were also performed in the absence of severe renal insufficiency or hemodynamic instability. Data were electronically stored and used as noted at the time of the original examination without alteration.
Patients were treated according to each center's local recommendations and followed for 6 months after the initiation of antibiotic therapy. Follow-up was obtained through clinical records and telephone calls to patients and their physicians. The primary endpoint was symptomatic embolic events that occurred during the 6-month period after the initiation of antibiotic therapy and before valve surgery. These events were defined as sudden clinical symptoms of cerebral ischemia or peripheral or pulmonary embolism. Peripheral and pulmonary emboli were systematically confirmed by computed tomography, and/or magnetic resonance imaging, and/or lung ventilation-perfusion scintigraphy, and/or arteriography. A specific diagnosis of cerebral embolism was confirmed by an experienced neurologist and by imaging. Cutaneous manifestations were not included. Data on valve surgery or death were also collected.
First, a descriptive analysis of recorded data was performed among the total cohort. Continuous variables are expressed as mean ± SD or median (interquartile range), and categorical data are expressed as number (percent).
Then, the prognostic model for ER was constructed using the following procedure: 1) the total cohort was randomized into 2 groups, resulting in development (two-thirds of the total cohort) and validation (one-third of the total cohort) samples; and 2) a prognostic model was constructed and validated on the basis of the development and validation samples, respectively. However, in this clinical context, the occurrence of embolic events may be altered or precluded by cardiac surgery or death, regardless of the cause, creating a context of competing risks (20). Therefore, our analysis accounted for this potential bias, as previously described (21). The analysis was limited to the first occurring event in the competing risks framework using the following quantities commonly used to summarize outcomes by event type: 1) the cause-specific hazard function, which for embolic events can heuristically be thought of as the probability of embolism in a small interval of time, given that no cardiac surgery or death occurred before; and 2) the cumulative incidence function, which for embolic events corresponds to the probability of embolism in the presence of competing cardiac surgery or death events. The delay from the initiation of antibiotic therapy to the first observed event was calculated in each case, and follow-up was restricted to the first 6 months after the initiation of antibiotic therapy, a time when patients still at risk were censored.
More precisely, patients were randomly assigned to 1 of the samples according to the type of event of interest (i.e., embolism, cardiac surgery, and death) and the medical center. Clinical and pre-clinical characteristics of the patients were described for the 2 samples and compared. For qualitative variables, chi-square tests were performed when valid, and Fisher exact tests were performed otherwise. For quantitative variables, Student t tests were performed when valid, and Mann-Whitney tests were performed otherwise. To assess the quality of the randomization, we estimated the cumulative incidence of embolic events, accounting for competing risks, and we compared these incidences according to the samples (development or validation) using a Gray test (22). To construct the prognostic model for embolic events, univariate analyses were first performed with the development sample to identify prognostic factors of embolic events. The following variables, established at diagnosis, were tested as potentially predictive of ER: age, sex, diabetes, comorbidity index, atrial fibrillation, previous embolism, heart failure, aspirin, anticoagulant therapy, valve localization, presence of vegetation, vegetation length, periannular complications, left ventricular ejection fraction, causative microorganism, and calendar year. Subdistribution hazard ratios and their 95% confidence intervals (CIs) were estimated using the Fine and Gray model (23). The proportional hazards assumption was assessed by testing covariate interactions with quadratic function of time and checked graphically using Schoenfeld-type residuals (24). All variables with p values <0.15 were included in the multivariate model to compute the ER. Their effect was studied in the multivariate model, but no variable selection was performed. The predictive accuracy of the prognostic model was assessed by studying calibration (agreement between predicted and observed risks) and discrimination (adapted C-index [21,25] and Royston and Sauerbrei's measure D  with bootstrap 95% CI). Moreover, in the development and validation samples, patients were classified according to the median of the distribution of the predicted probability of the occurrence of embolic events. Then, cumulative incidences for embolic events were estimated in each of these subgroups and compared using the Gray test (22).
All of the tests were 2-sided. A p value <0.05 was considered to be significant. Statistical analysis was performed using R version 2.14.0 (R Foundation for Statistical Computing, Vienna, Austria). The R package cmprsk was used for competing risk analyses.
During the study period, 1,022 patients with definite diagnoses of infective endocarditis were treated at the 2 centers. Among these, 135 were excluded because of endocarditis episodes that involved only a pacemaker or defibrillator lead. An additional 40 patients were excluded because of previous endocarditis episodes for which they were already included. Data are thus presented for 847 patients. The mean age of the patients was 62 ± 15 years, and 71.5% were men. Streptococci were the most frequent causative pathogens (33.6%), and Staphylococcus aureus was the cause in 17.1% of the patients. In total, 226 patients (26.7%) had prosthetic valve endocarditis. Computed tomography could be performed at admission in 823 patients (97%). Of the 318 patients with previous embolism, 134 had symptomatic events, 217 had silent events, and 33 had both symptomatic and silent events. Four hundred ninety-three patients underwent valve surgery after a median time of 10 days (interquartile range: 3 to 27 days) after the beginning of antibiotic therapy. Among them, the indications of surgery were acute heart failure in 227 (46%), periannular complications in 172 (35%), and large vegetations with or without previous embolism in 185 (36%). Eighty-nine patients (18% of operated patients and 11% of all patients) were operated only on the basis of the vegetation length, with or without previous embolism. During the study period, there was no significant modification of the overall rate of surgery (p for trend = 0.09) and the rate of surgery for a high ER (p for trend = 0.99).
The total cohort was randomly divided into the development sample (n = 565) and the validation sample (n = 282). The 2 samples were similar with respect to the main baseline characteristics (Table 1).
Incidence of embolic events according to competing risks
In the total cohort, embolic events occurred in 72 patients a median of 6.5 days (interquartile range: 3 to 14 days) after the initiation of antibiotic therapy. The sites of embolization were the central nervous system (n = 36), peripheral arteries (n = 16), spleen (n = 10), kidneys (n = 4), eyes (n = 2), coronary circulation (n = 2), and mesenteric circulation (n = 2). At 6 months, valve surgery was performed in 61.7% of patients (95% CI: 58.1% to 65%), and 20% of patients (95% CI: 17.2% to 22.7%) died.
Accounting for the competing risks, the 6-month cumulative incidence of embolic events was 8.5% (95% CI: 6.7% to 10.4%) in the total cohort and was similar in the development and validation samples (p = 0.99). The incidence of embolic events was highest during the first 2 weeks after the initiation of antibiotic therapy (44.9 embolic events per 1,000 patient-weeks in the first week and 21.3 embolic events per 1,000 patient-weeks in the second week) and then decreased rapidly and significantly to low in the sixth week (2.4 embolic events per 1,000 patient-weeks) (Fig. 1).
The ER prediction model
In the development sample, the variables associated with embolic events were age, diabetes, atrial fibrillation, previous embolism, vegetation length, and Staphylococcus aureus. Using these variables, a multivariate ER prediction model was developed (Table 2). The calibration of the model was excellent, as shown by the excellent correlation between the predicted and observed risk for embolic events (Fig. 2A). The predictive accuracy of this prediction model was excellent, with a C-index of 0.72 (95% CI: 0.65 to 0.81), Royston and Sauerbrei's measure D of 1.26 (95% CI: 0.81 to 1.70), and a significantly higher cumulative incidence of embolic events observed in patients with high predicted ER (p < 0.0001) (Fig. 3A).
The good predictive accuracy of the model was retained when it was tested in the validation sample. In this sample, there was also an excellent correlation between the predicted and observed ER (Fig. 2B). Moreover, the C-index was 0.65 (95% CI: 0.55 to 0.77), Royston and Sauerbrei's measure D was 0.82 (95% CI: 0.22 to 1.41), and there was a significantly higher cumulative incidence of embolic events in patients with high predicted ER (p < 0.05) (Fig. 3B).
After the exclusion of patients who already had reasonable indications for early surgery (severe valvular regurgitations and/or periannular complications), all patients remained correctly classified by the calculator (p = 0.01) (Fig. 4).
Figure 5 illustrates the method to calculate a patient's ER. This model can be programmed into a handheld device to make risk calculation automatic once the individual variables have been entered (Online Appendix).
We have developed and validated a simple bedside prediction system that can be used to quantify the ER at admission of patients with endocarditis. By using a large multicenter cohort and focusing on clinically simple and relevant variables, we believe that this tool will be usable and reliable for clinicians and will help them make treatment decisions.
A recent randomized trial demonstrated the benefit of early surgery on the risk for embolism. Although this result is of crucial importance, it was limited by the fact that it was obtained in a population with very low operative risk. Thus, we now have strong evidence that early surgery reduces ER, but there remains a need for better risk stratification to evaluate accurately the benefit/risk ratio of this procedure. Indeed, for high ER associated with low or intermediate predicted operative mortality (computed by scoring systems ), the benefit of early surgery would be greater.
The vegetation length and the presence of a previous embolism are the only 2 parameters included in the present international guidelines to assess ER and indicate preventive valve surgery (13,14). These recommendations do not provide a precise quantification of ER and do not take into account other potentially important predictors. Our embolic prediction model is the only current method to accurately quantify this ER. This model is robust because it was developed in a large multicenter sample and validated in an independent second sample. Moreover, it takes into account predictors both related to the disease and associated with patient characteristics.
We demonstrated that vegetation length and the presence of Staphylococcus aureus were significantly and nearly significantly, respectively, associated with ER, as previously found (11,12). These results emphasize the role of echocardiography in the prediction of embolisms and the absolute necessity of a precise and fast microbiological diagnosis (1). Although vegetation mobility was previously demonstrated to be a potential predictor of ER, this variable was not tested in our model because the interobserver variability of its evaluation is known to be low (28). Indeed, vegetation mobility is not a criterion recommended in the international guidelines to indicate surgery. The occurrence of an embolism before medical therapy was included in our prediction model. According to previous studies, we confirmed the impact of silent embolism screening in the prediction of ER (5,11). Systematic computed tomographic scanning was used to detect these asymptomatic embolic events but was limited by its low sensitivity for small cerebral damage and the risk for renal failure related to iodine injection. More sensitive imaging modalities, such as magnetic resonance imaging (3,5) and 18-fluorine fluorodeoxyglucose positron emission tomography, could be relevant alternatives to improve our predictive model (29,30).
In addition, our prediction model included variables that were not related to the disease itself but to prothrombogenic conditions directly related to patient characteristics, namely, advanced age, diabetes, and atrial fibrillation. Indeed, these are conditions that may influence the formation and adhesion of thrombi to endothelial cells and influence ER (31,32). Several studies have demonstrated that host humoral factors, such as those involved in the coagulation and fibrinolysis phenomena, are additional determinants of ER during infective endocarditis (6,7,33,34). Although the influence of these factors on ER is lower than that of vegetation length, their presence associated with large vegetation would increase the risk for embolism and thus could influence the decision to remove surgically the main risk factor of embolism, namely, the vegetation. Moreover, future medical therapeutic strategies could be developed to decrease the global thrombotic activity involved in the pathophysiology of endocarditis.
The incidence of embolic events after the initiation of medical therapy was estimated to be from 6% to 21% in previous studies (10,14). This heterogeneity is due mainly to differences in selection criteria and in the frequency of valve surgery, which vary from center to center. This therapeutic procedure has never been appropriately addressed in the assessment of ER. Thus, we considered this bias using a statistical method that accounted for competing risks (i.e., valve surgery and death), which may preclude the occurrence of embolic events. Thus, we demonstrated that the incidence of embolic events was the highest during the first 2 weeks after the initiation of antibiotic therapy and decreased thereafter. This observation most likely illustrates the beneficial effect of antibiotics on ER by modifying the biological constitution of vegetations and confirms the previous uncontrolled studies (8). Therefore, our calculator should be used as soon as possible after diagnosis so that valve surgery can be performed in an urgent setting if ER is high (10). However, further studies will be needed to determine at which level of ER a significant benefit of valve surgery can be observed. Nevertheless, this tool will help clinicians in the standardization of endocarditis management (35). Moreover, it could be used in future studies to investigate the impact of new treatment strategies on the occurrence of embolic events.
First, the 2 samples were assembled retrospectively; thus, the study was exposed to a detection bias for the determination of the baseline characteristics. However, we have been developing a common database for several years that includes consecutive patients, with rigorous definitions of variables collected.
Second, in our prediction model, we considered only the predictors evaluated at admission, without taking into account their modifications after the initial evaluation. However, because the objective of the study was to provide support for rapid therapeutic decisions to avoid embolic events, data collected at admission were the most important.
Third, left atrial dimensions and glycosylated hemoglobin are parameters that could influence the prothrombotic state but were not collected in the present study.
Fourth, the relatively small number of events could create the risk for overfitting the model when adding covariates. However, we tested only the few clinically relevant variables known to be potentially important predictors of the outcome. This strategy could decrease the probability to include predictors by chance.
Fifth, although the predictive properties of proposed model was assessed in an internal validation step, it needs to be assessed in external data to increase its generalizability.
Finally, the study was subject to a referral bias because it was performed at referral centers.
The evaluation of ER is crucial in the management of infective endocarditis. We thus developed and validated a new prediction system to quantify the ER at admission: the Embolic Risk French Calculator (Online Appendix), which can be used by clinicians from a dedicated Web site or as a program installed on a handheld device. This risk index may be useful for facilitating management decisions and may allow future researchers to address critical and difficult problems about weighing the risks and benefits of early surgery in individual patients with infective endocarditis.
The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- embolic risk
- Received May 18, 2013.
- Revision received July 4, 2013.
- Accepted July 12, 2013.
- American College of Cardiology Foundation
- Thuny F.,
- Avierinos J.F.,
- Tribouilloy C.,
- et al.
- Cooper H.A.,
- Thompson E.C.,
- Laureno R.,
- et al.
- Snygg-Martin U.,
- Gustafsson L.,
- Rosengren L.,
- et al.
- Kupferwasser L.I.,
- Hafner G.,
- Mohr-Kahaly S.,
- et al.
- Kim D.H.,
- Kang D.H.,
- Lee M.Z.,
- et al.
- Vilacosta I.,
- Graupner C.,
- San Roman J.A.,
- et al.
- Thuny F.,
- Di Salvo G.,
- Belliard O.,
- et al.
- Baddour L.M.,
- Wilson W.R.,
- Bayer A.S.,
- et al.
- Habib G.,
- Hoen B.,
- Tornos P.,
- et al.
- Selton-Suty C.,
- Celard M.,
- Le Moing V.,
- et al.
- Li J.S.,
- Sexton D.J.,
- Mick N.,
- et al.
- Camm A.J.,
- Lip G.Y.,
- De Caterina R.,
- et al.
- Nashef S.A.,
- Roques F.,
- Michel P.,
- et al.
- Di Salvo G.,
- Habib G.,
- Pergola V.,
- et al.
- Feinbloom D.,
- Bauer K.A.
- Korkmaz S.,
- Ileri M.,
- Hisar I.,
- et al.