Mortality rate of left-sided IE remains high. Although rates lower than 20% have been reported [7, 22, 23], they usually range between 20 and 30%[4, 5] and are even higher in some subgroups; mortality rates of 50% has been reported in early prosthetic valve IE due to S. aureus . Crude mortality rate in our series was 29.5%, an intermediate figure considering the intrinsic features of the patients, the number of complicated cases, and the length of the study period. Considering the evolution of mortality over time, we observed a non-significant upward trend, probably related to an increased percentage of cases caused by more virulent organisms such as S. aureus, and to a higher frequency of elderly patients with comorbidities.
Previous studies about the prognosis of IE have often been incomplete in their collection of risk factors, have included heterogeneous subtypes (ie, left-side and right-side endocarditis) or have been carried out on patients treated in a single reference center. Only few multicenter cohort with multivariate analysis studies including large numbers of patients have been reported [6, 7, 9, 10, 13, 25], of which one included only patients admitted to the ICU  and another only patients with prosthetic valve IE . This prospective multicenter study includes cases from both referral and community hospitals. We also include IE in both prosthetic and native valves and we have included most factors found in previous studies, such as patient-related variables, aetiology, and the development of different complications.
A Charlson index of more than 2 points has been reported to be an independent risk factor for mortality in previous studies [6, 9]. We used the age-adjusted index on the grounds that age may be related to prognosis  and also found it to be related with increased risk of death. Many of our patients had more than one underlying condition. Insulin-dependent diabetes mellitus has been observed to be a risk factor in one study  but we did not find such association in ours. On the other hand, a very high mortality rate was found among patients with chronic liver disease (47%), similarly to what has been reported in another Spanish series .
Prosthetic valve IE, particularly early episodes and those caused by S. aureus, are usually associated with a worse prognosis than native valve IE [23, 29]. Crude mortality of prosthetic valve IE in our series was almost 3 times higher than for native IE, corroborating the results of previous studies [12, 29, 30]. Furthermore, S. aureus was independently associated with increased mortality, whilst Streptoccocus viridans group was a protective factor. Our results confirm another recent collaborative multicenter study where mortality for IE was higher when due to S. aureus (22.4% vs 14.6% for other aetiologies); association of this microorganism with life-threatening complications, such as septic shock and neurologic manifestations have also been reported . All these data emphasise the prognostic relevance of this microorganism in IE.
Heart failure has been found to be a strong marker for increased mortality in most studies . In contrast to Chu et al , we did not find heart failure at admission to be associated with a worse prognosis. A possible explanation for this is that heart failure at admission, when due to treatable disorders such as arrhythmia, fluid overload, etc, may be controlled by medical treatment. On the other hand, the development of moderate to severe heart failure during evolution was related to increased mortality. We think that development of heart failure during evolution is probably due in most cases to progressive valve dysfunction, which would only be modifiable by surgery. In fact, surgery was a protective factor in our study. Surgical treatment was a protective factor in only two previous studies [8, 9], particularly in complicated cases with severe heart failure . The fact that surgery is an independent protective factor for mortality in our study is possibly related to the high number of complicated cases included with heart failure due to valvular dysfunction. However this date must be interpreted with caution since this is an observational study; in another recent study , surgery during episode of IE either was associated with better prognosis when this variable was analysed in a multivariable regression modelling.
The exact impact of surgical treatment in IE is controversial and difficult to assess, since evidence is based in observational studies but not in clinical trials. Some new methods of analysis, such as the use of propensity score, have been applied in some studies in order to improve the control of confounding. However the results of these studies are conflicting probably because of methodological differences, as has been recently reviewed .
Neurological complications are frequent and associated with a worse prognosis [7, 10, 12]. In this study, as found in others, about one quarter of patients developed neurologic complications . We considered this complications only when it was clinically relevant, and not only like alterations in imaging techniques. Embolic phenomena are known to be more frequent at admission and during the first week, to occur despite proper medical treatment, and to be associated with worse prognosis, as was the case in our series.
Septic shock, which was developed by about 10% of our patients, was independently associated with increased mortality. This variable has only been studied previously in a report detailing IE patients admitted to the ICU , probably because it is not a frequent complication and cannot be evaluated as an independent risk factor in series with a low number of cases. As septic shock has been associated in general patients with mortality rate of about 50% , we consider that this variable helps to identify a high-risk subgroup of patients that may benefit from aggressive management, such as early identification of severe sepsis and triggering evidence-based protocolised care .
Vegetations observed in echocardiography have been previously associated to a poor outcome  although one study found only large-sized vegetations to be an independent prognostic factor . In our study, a vegetation size of >20 mm almost reached statistical significance in univariate analysis but did not in multivariate analysis; a limitation of our study is that size of vegetations was not measured in all cases.
Perivalvular extension of infection (abscess and/or fistula) has traditionally been seen as an unfavourable prognostic factor [7, 37] although less so in more recent studies [25, 38]. As in another recent large study  we found this complication to be an independent risk factor for mortality, which may be related to a higher post-surgical mortality associated with the intrinsic technically difficulties of these cases. This is a variable that identify another high-risk subgroup of patients, and must be suspected in presence of risk factors for this complications like prosthetic IE and aortic localization .
Our study have some limitations: First, changes in the clinical management of IE may have occurred during the long study period, although the inclusion of this variable in the multivariate model did not affect the results. Second, the majority of cases were included from referral centers so that the percentage of complicated cases may have been overestimated. However, other multicenter studies only include cases from reference centers and, even if represented by a small percentage, our series probably better reflects the overall clinical reality. Third, as TEE was not systematically performed, evaluation of echocardiographic data (size of vegetations, perivalvular extension) may be biased. And fourth, it is an observational study, so some results, such as the protective effect of surgery, must be interpreted cautiously.