In-hospital mortality of infective endocarditis : Prognostic factors and evolution over an 8-year period

Infective endocarditis (IE) remains severe. Few predictors of prognosis have been identified. It is not known whether mortality of IE has decreased during recent decades. 559 definite cases of IE were collected in a prospective population-based survey in 1999 in France. In-hospital death rate was 17%. It was lower in operated patients (14.4% vs 19.3%), although not significantly so. In multivariate analysis, the following variables were independent and significant predictors of mortality: history of heart failure (odds ratio: 2.65), history of immunosuppression (OR: 3.34), insulin-requiring diabetes mellitus (OR: 7.82), left-sided IE (OR: 1.97), heart failure (OR: 2.19), septic shock (OR: 4.33), lower Glasgow coma scale score (OR: 4.09), cerebral haemorrhage (OR: 9.46), and higher C-reactive protein level (OR: 2.60). Adjusted mortality was significantly lower in 1999 than in 1991 (22%): OR: 0.64 (p=0.03). Thus, in a large and unselected cohort of patients hospitalized for IE in 1999, in-hospital mortality rate was lower than in 1991. Multivariate analysis identified factors classically known as having an impact on mortality. However, other factors, such as age and responsibility of Staphylococcus aureus, were not retained in the model.

Although infective endocarditis (IE) is relatively rare, it remains a severe disease, with a death rate of 12% to 20% during the initial hospital phase [1], [2], [3], [4], [5], [6], [7], [8] and an annual death rate of about 3% in survivors of the hospital phase [2], [9], [10], [11].

Several variables are often reported as markers of poor prognosis: older age, heart failure, renal failure, infection due to staphylococci, aortic location, embolisms, IE on prosthetic valve, and persisting fever under antibiotic treatment [2], [4], [9], [11], [12]. However, many studies on the prognosis of IE suffer from biases and limitations, such as single-centre recruitment (often in a tertiary-care hospital); retrospective studies; low number of cases; few potential predictors of prognosis studied; the list of potential prognostic factors differing from one study to another; multivariate analyses seldom performed. We thus undertook a prospective study of in-hospital mortality in a large number of patients with IE, recruited from all types of hospital settings (private and public hospitals, primary, secondary and tertiary hospitals). We explored the potential role of many variables, both long-established prognostic factors and newer variables, using both univariate and multivariate analyses.

It is not known whether mortality of IE has decreased over time during recent decades, since comparative surveys properly carried out in the same defined areas using similar methods are lacking. We thus compared the results of this 1999 survey with those of a similar study conducted by the same group in 1991 [1].

Cases of IE were collected during a cross-sectional prospective population-based survey that was conducted in 1999 in France. Methods and results of this survey have been published previously [13]. The survey was publicized by mail to all physicians working in hospitals of these regions who were either physicians likely to take care of patients with IE (i.e. infectious diseases, intensive care, and internal medicine physicians, cardiologists, and cardiac surgeons), echocardiographists, or microbiologists. Physicians were asked to fax a notification form to the study-coordinating centre for each suspected case of IE. They were reminded of this study on a regular basis by mail. For each patient older than 15 y of age and living in 1 of the study regions, specific case report forms were sent to both the attending physician and the microbiologist. The study complies with the Declaration of Helsinki and was approved by the Commission Nationale Informatique et Liberté (National Computing and Freedom Committee). The survey was launched on 1 December 1998 and stopped on 31 March 2000. During this time frame, 653 cases of IE were entered in the database. All case report forms were checked and validated by 2 expert investigators who had not been involved in the care of the corresponding cases. These investigators were responsible for validating the diagnosis of IE according to the Duke criteria [14]. This process led to exclusion of 94 cases and retention of 559 definite IE cases, 390 of which were retained for case description and 1-y incidence calculation in the original manuscript because their hospitalization date fell between 1 January and 31 December 1999 [13].

In the present study, in order to increase the power of the analysis, all 559 definite cases of IE were analysed. The original 390 cases and the 169 additional ones were compared for age, gender, pre-existing valve disease, location of IE, causative microorganism, rate of surgical treatment, and death rate. No statistically significant difference was evident between the 2 groups (data not shown), which validates the decision to pool the 2 groups.

The list of studied variables is presented in Table I. Prognostic influence of variables on in-hospital mortality was tested first in univariate analysis (Pearson's χ2 test), then in multivariate analysis (stepwise logistic regression, variables with p<0.10 in univariate analyses). We ran 3 models, 1 with baseline variables only, 1 with baseline variables 'forced' to remain in the model and 'evolution' variables (detailed in Table I) and 1 with baseline variables not forced and 'evolution' variables. For each model, we provided the area under the ROC curve given by the c statistic (null prediction: c statistic = 0; perfect prediction: c statistic = 1). All analyses were performed using SAS software.

Table I.  Studied variables.

In order to understand why some classical risk factors for higher death rate (age, Staphylococcus aureus) or for lower death rate (surgery) were not significant in the multivariate models, we looked at the correlations between these variables and all other variables (results in appendix).

Comparison of the 1991 and 1999 databases was restricted to patients older than 18 y, and fulfilling the Beth Israel criteria modified with echocardiography for definite, probable or possible IE [1]. The latter were used because the Duke criteria were not available in 1991 and could not be applied retrospectively. Mortality rates were adjusted for age, gender, location of IE and microorganism.

The main characteristics of the population are presented in Table II. Briefly, mean age was 59.0±16.8 y, and 72% of the patients were male. Almost half of the patients had no previously known heart disease, one-third had native valve disease, and 15% had prosthetic valve. Causative microorganisms were Streptococcaceae in more than half of the cases, Staphylococcaceae in almost a third; no microorganism was identified in 5% of the cases.

Table II.  Description of the population.

In-hospital death rate was 17% (95/559). Moment of death is displayed in Figure 1 The rate of valve surgery during the initial hospital stay was 47% (264/559).

Graph: Figure 1. Timing of death and of cardiac surgery (number of patients).

After the 1991 and 1999 series were made comparable, death rates were 21.6% in 1991 and 16.6% in 1999 (p=0.08). Since the 2 populations were different (64% vs 71% men, 24% vs 33% patients >70 y of age, 90% vs 78% left-sided IE, and 16% vs 21% Staphylococcus aureus IE, respectively, in 1991 and 1999), we adjusted the 2 populations on age, gender, location of IE and causative microorganism. After adjustment, mortality was significantly lower in 1999 than in 1991 (OR 0.64: 95% CI 0.43–0.96; p=0.03).

Univariate prognosis analysis for in-hospital mortality is presented in Table III. Several factors often cited as associated with higher in-hospital mortality – older age, history of prosthetic valve, heart failure (New York Heart Association-NYHA-class III or IV, need for diuretic treatment), septic shock, low Glasgow coma scale score, renal failure, cerebral haemorrhage, elevated C-reactive protein, Staphylococcus aureus – were also significant predictors in our study. There was a non-significant trend towards lower death rate in right-sided IE. Echocardiography variables (vegetation, abscess, valve regurgitation, prosthesis dehiscence) did not contribute significantly to prognosis. The death rate was lower (14.4%) in operated patients than in non-operated patients (19.3%), although the difference was not statistically significant (p=0.12).

Table III.  Univariate analysis of death (only variables for which p<0.10).

703 p>0.10: baseline variables: gender, non-insulin-requiring diabetes mellitus, history of high blood pressure, dyslipidaemia, history of coronary heart disease, of lower limb arterial disease, of cerebrovascular accident, of autoimmune disease, of psoriasis, of hepatic disease, pregnancy, history of native valve disease, of heart murmur, of pacemaker, of infective endocarditis, at-risk procedure, at-risk situation, i.v. drug use, vegetation, abscess, valve regurgitation, prosthesis dehiscence, surgery; 'evolution' variables: embolisms, mycotic aneurysm, petechiae, Janeway erythema, glomerulonephritis, Roth spots, Osler nodes, meningitis, white blood cells count >10 109/l, fever >38°C, secondary location. Statistical test: χ2 or Fisher's test according to the size of the cells.

The first multivariate model included baseline variables only (Table IV). History of heart failure, history of immunosuppression, insulin-requiring diabetes mellitus, older age, left-sided IE, and Staphylococcus aureus were associated with a higher mortality. The c statistic was 0.742. We also ran 2 models that assessed both baseline and evolution variables. In a first step, baseline variables that were significant in the previous model were forced. The following evolution variables were significant predictors of a higher mortality: heart failure, septic shock, lower Glasgow coma scale score, cerebral haemorrhage, and higher C-reactive protein level. The c statistic was 0.843. In a second step, baseline variables were not forced; older age and Staphylococcus aureus did not remain in the model. The c statistic was 0.837. Goodness of fit of these models was confirmed by Hosmer and Lemeshow test p-values of 0.63, 0.81, and 0.64, respectively).

Table IV.  Factors independently associated with in-hospital mortality (stepwise logistic regression).

Mortality was lower in operated patients, but this was not significant (OR: 0.723, 95% CI: 0.388–1.349).

Mortality rate during the initial phase of IE was 17% in our study. This rate is within the range reported in recent series [1], [2], [3], [6], [7], [8], [10], [15], [16], [17], [18], [19], [20].

Many factors that have been identified as prognostic factors in these series were prognostic factors in our study, at least in univariate analysis. These were age, history of prosthetic valve, left-sided IE, heart failure, signs of severe infection (C-reactive protein >120 mg/l, septic shock), Staphylococcus aureus IE, cerebral haemorrhage, renal failure. Embolism was not a prognostic factor. Although it is often regarded as a prognostic factor in older studies, this fact was not confirmed in the most recent studies [17]. At the same time, our study disclosed some prognostic factors that were not known, most probably because they were not looked at in other studies. These were history of insulin-requiring diabetes mellitus, heart failure, respiratory insufficiency, chronic renal failure, cancer, or immunosuppression, as well as obesity and low Glasgow coma scale score. The prediction of in-hospital mortality by our multivariate model was excellent, as shown by a c statistic value of 0.84.

There are only few studies with multivariate analysis of prognosis in IE [6], [8], [17], [19], [20]. The first seems to be the study by Woo [17]. In this now old study, among about 20 variables, the authors retained 4 to express the prognostic index: leukocytosis (no cut-off value given), major systemic embolism, heart failure and some microorganisms (staphylococci, Pseudomonas, Klebsiella and ß-haemolytic streptococci).

Netzer retrospectively evaluated 212 patients who were hospitalized between 1980 and 1995 in 1 tertiary-care centre [6]. Among about 20 variables, 5 remained in the multivariate model. Three variables were associated with an increased risk of death – presence of neurological symptoms on admission (OR: 26.1), arthralgia (6.2) and weight loss (4.2). Two variables were associated with a decreased risk of death – higher number of symptoms on admission and presence of non-cardiac risk factors or underlying conditions (i.v. drug use, HIV, creatinine concentration >110 µmol/l, diabetes mellitus, pregnancy and cancer) (ORs not given). This may be due to the fact that these 2 variables led the treating physician to make the diagnosis and to start treatment earlier. Prosthetic valve IE, serum creatinine, valve location, heart failure and surgery had no prognostic significance. In 159 patients hospitalized between 1990 and 1993 in a single institution, heart failure and renal failure were independently associated with increased mortality in multivariate analysis [19]. Of note, in this study 67% of the patients were i.v. drug users.

In a retrospective study of 208 patients hospitalized in a teaching hospital between 1981 and 1999, only 2 of about 35 variables remained in the multivariate model: an abnormally low (<3×109/l) or high (>11×109/l) WBC increased the risk of death by almost 9 times; a low serum albumin concentration (<30 g/l) increased the risk by almost 5 times [8]. This is the only study that establishes low albumin concentration as a pejorative factor in IE, but it is known as a prognostic factor in various states of sepsis [21], [22]. Unfortunately, in our study we did not look for the influence of albumin.

Thus, in all these studies, only very few predictive factors remained in the multivariate analysis (2 to 5), although many variables were studied (20 to 35). The list of variables differed from 1 study to another. While some classical risk factors were significant in some studies, these studies also disclosed unusual factors.

Our analysis also brings some surprising results: while heart failure, septic shock, low Glasgow coma scale score, and cerebral haemorrhage remained in the multivariate model, age and Staphylococcus aureus did not. These 2 classically predictive factors entered the model including baseline variables only, but they were omitted when evolution variables were taken into account. Although surgery was associated with a lower mortality rate (14.4% vs 19.3%), it was not statistically significant in univariate analysis, and surgery did not remain as a factor of better prognosis in the multivariate model (OR: 0.723, 95% CI 0.388–1.349).

Because of these findings, we looked at the relationships between age, Staphylococcus aureus and surgery on the 1 hand and all other variables in the other. The fact that these variables did not remain in multivariate analysis can be explained in part by their link with several variables that are potent predictors of in-hospital mortality. Age was linked with history of heart failure and present heart failure. In most recent studies using multivariate analysis, age was not a prognostic factor either [6], [8], [17]. Staphylococcus aureus was linked with septic shock, low Glasgow coma scale score, cerebral haemorrhage and very high C-reactive protein level. Again, Staphylococcus aureus is not an independent predictor in some recent multivariate analyses [6], [8].

Surgery was not an independent predictor of a lower mortality, but it was strongly linked with heart failure, systemic embolism, severe lesions at echocardiography (abscess, prosthesis dehiscence), all features that are classical indications for early surgery. It is often written that surgery decreases the mortality rate of IE. However, surgery is performed in complicated cases only. Again, surgery does not emerge as a significant factor in recent studies [6], [8], [17], [19], [20].

The in-hospital mortality rate of IE has decreased over time; when the 2 French series are made comparable, the death rate was 21.6% in 1991 and 16.6% in 1999. After adjustment on age, gender, location of IE and microorganism, mortality significantly decreased by a third (OR: 0.64) between 1991 and 1999 (p=0.03). To our knowledge, this is the only study that allows looking at the evolution of mortality over time. We eagerly await other studies in other countries in order to confirm a decrease of early mortality of IE.

Since we adjusted on age, location of IE and microorganisms, these 3 variables cannot explain the decrease of mortality. Among possible explanations (better medical care, newer antibiotics ...), the most probable is that recent patients were operated much more often (50% vs 31%; p<2.10−7). The fact that surgery is not a significant predictor of lower mortality in the multivariate analysis of the 1999 dataset does not militate against the fact that surgery explains a decrease of mortality over time.

In conclusion, in a large and unselected series of patients hospitalized for IE in 1999, the in-hospital mortality rate remained high (17%). It has decreased over time, since this rate was 22% in a similar study performed in patients hospitalized in 1991. The multivariate analysis of mortality disclosed factors classically known as influencing mortality, such as heart failure, septic shock, cerebral haemorrhage. However, other often cited factors did not remain in the model – age and Staphylococcus aureus, most probably because of their strong association with the previously cited factors.

The study was funded by the Programme Hospitalier de Recherche Clinique (Grant PHRC 1997: RC30), the Fédération Française de Cardiologie, and the Aventis and GlaxoSmithKline laboratories, France.

Association pour l'Etude et la Prévention de l'Endocardite infectieuse (AEPEI) Study Group: Steering committee: B. Hoen, C. Leport (principal investigators), S. Briançon, N. Danchin, F. Delahaye, J. Etienne (co-investigators). Region study coordinators: Franche-Comté: Y. Bernard, F. Duchêne, P. Plésiat; Lorraine: F. Alla, T. Doco-Lecompte, M. Weber; Marne: I. Béguinot, P. Nazeyrollas, V. Vernet; New Caledonia: B. Garin, F. Lacassin, J. Robert; Ile-de-France: A. Andremont, E. Garbarz, V. Le Moing, J.L. Mainardi, R. Ruimy; Rhône-Alpes: C. Chidiac, F. Delahaye, F. Vandenesch. Other members: A. Bouvet, P. Bruneval, X. Duval, V. Goulet, R. Roudaut, R. Salamon, J. Texier-Maugein. Research assistants: S. Boucherit, Y. Bourezane, W. Nouioua, D. Renaud. Centre National de Référence des Streptocoques: A. Bouvet, G. Collobert, B. Merad, L. Schlegel. Centre National de Référence des Toxémies à Staphylocoques: M. Bes, J. Etienne, F. Vandenesch. Centre National de Référence des Legionella: J. Etienne, S. Jarraud, M. Reyrolle, Centre National de Référence des Rickettsies: J.P. Casalta, D. Raoult.

This work was supported by the Fédération Française de Cardiologie and the following professional organizations: Association des Professeurs de Pathologie Infectieuse et Tropicale, Société de Pathologie Infectieuse de Langue Française, Société Française de Microbiologie, Société Nationale Française de Médecine Interne, Société de Réanimation de Langue Française, Société Française de Gérontologie, Société Française de Cardiologie, Société Française de Chirurgie Thoracique et Cardiovasculaire, Société Française d'Anesthésie-Réanimation, Ligue Française pour la Prévention des Maladies Infectieuses. The authors are fully indebted to the physicians and microbiologists who participated in this survey.

The experiment complies with French current laws.

Relationships of age, Staphylococcus aureus, and surgery with other variables.

• 251 (45%) patients were <60 y old, 128 (23%) were 60–70 y old, 143 (25%) were 70–80 y old, and 37 (7%) were >80 y old. In univariate analysis, the following relationships were statistically significant: women were older than men; older patients more often had a history of high blood pressure, coronary heart disease, heart failure, cerebrovascular accident, cancer, valve disease, and pacemaker. Older patients more often developed NYHA class III or IV heart failure and vascular phenomena. They less often smoked, had a history of hepatic disease, pulmonary embolism, tricuspid valve location. All the 30 i.v. drug users were <60 y old. Staphylococcus aureus was less frequently involved in patients 60–70 y old (12%) than in the other patients: <60 y, 27%; >70 y, 26%. Older patients were operated on less often: surgery rate: <60 y, 55%; 60–70 y, 50%; 70–80 y, 41%; >80 y, 5% (p < 0.0001). They died more often: <60 y, 11%; 60–70 y, 18%; >70 y, 24% (p=0.004).
• 120 IE were due to Staphylococcus aureus. In univariate analysis, the following relationships were statistically significant: Staphylococcus aureus-IE was more frequent in younger patients. It was more frequent in insulin-dependent diabetic patients, in patients with a history of cerebrovascular accident, of chronic renal failure, of hepatic disease, of autoimmune disease. Patients with Staphylococcus aureus-IE more often had no previously known heart disease. In Staphylococcus aureus-IE, the Glasgow coma scale score was lower, septic shock occurred more often, as fever, vascular phenomena, purpura, glomerulonephritis, meningitis, pulmonary embolism, serum creatinine level >180 µmol/l, white blood cells count (WBC) >10×109/l, C-reactive protein >120 mg/l. In Staphylococcus aureus-IE, the tricuspid valve was more often involved, whereas the aortic valve was less often involved than in IE due to other microorganisms. There was more often no or only 1 valve injured by IE. Vegetations were more frequent, whereas abscesses were less frequent. Patients with Staphylococcus aureus-IE were operated less often (31% vs 52%). They died more often (27% vs 14%). In multivariate analysis, Staphylococcus aureus remained a significant factor of a lower surgery rate (OR: 0.48), but it did not influence the mortality rate.

With regard to surgery, the following relationships were statistically significant in univariate analysis: older patients were operated on less often, as were patients with cerebral haemorrhage, those with pulmonary embolism, those with pacemaker, Staphylococcus aureus IE and tricuspid location. Patients with known valve disease were operated more often, as were patients in NYHA class III or IV, those with systemic embolism, those with more than 1 valve involved, mitral location, aortic location, regurgitation at echocardiography, abscess, and prosthesis dehiscence.