Article Text
PDF
Abstract
Introduction Early identification of prognostic factors is essential to improve the grim prognosis associated with left-sided infective endocarditis. This group identified three independent risk factors obtained within 72 h of admission, (Staphylococcus aureus, heart failure and periannular complications) for inhospital mortality or urgent surgery in a series of 317 patients diagnosed at five tertiary centres (derivation sample). A stratification score was constructed for the test cohort by a simple arithmetic sum of the number of variables present. The goal was to validate this model internally and externally in a prospective manner with two different cohorts of patients.
Methods The appropriateness of the model was tested prospectively on predicting events in two cohorts of patients with left-sided endocarditis: internally with the 263 consecutive patients diagnosed at the same centres where the model was derived (internal validation sample), and externally with 264 patients admitted at another hospital (external validation sample).
Results The discriminatory power of the model, expressed as the area under the receiver operating characteristic curve was similar between derivation and both validation samples (internal 0.67 vs 0.68, p=0.79; external 0.67 vs p=0.74, p=0.09). There was a progressive, significant pattern of increasing event rates as the risk stratification score increased in both validation cohorts (p<0.001 by χ2 for trend).
Conclusions The early risk stratification model derived, based on variables obtained within 72 h of admission, is applicable to different populations with left-sided endocarditis. A simple bedside assessment tool is provided to clinicians that identifies patients at high risk of having an adverse event.
- Endocardial disease
- endocarditis
- prognosis stratification
- risk score
Statistics from Altmetric.com
Despite the experimental advances in the diagnosis and treatment of left-sided infective endocarditis, it remains a high-mortality disease with steady percentages of mortality in the past 30 years.1–6 From a practical point of view, it would be very useful to identify a subgroup of patients with high risk of complications based on clinical, echocardiographic and microbiological variables obtained in the early stages of the disease. In this respect, our group identified three independent prognostic factors of inhospital mortality or urgent surgery available within 72 h of admission: Staphylococcus aureus, heart failure and periannular complications, and derived a stratification model based on the number and type of variables present to predict the probability of an event.7
The baseline characteristics and the management of patients with endocarditis differ profoundly among institutions, as was observed in the Euro Heart Survey;8 therefore, it is mandatory to test the appropriateness of our results in different populations. The generalisability of our prognostic model requires that the model' predictions remain reliable and accurately discriminate key outcomes when re-tested in other series of patients. In this study we internally and externally validate our stratification model in two different comparable cohorts of patients with left-sided infective endocarditis.
Methods
Initial derivation of the prognostic model
We derived our early stratification model from 317 adult patients consecutively diagnosed with infective endocarditis at five university-affiliated tertiary care hospitals with cardiovascular surgical facilities, referral centres for their regions on infective endocarditis from January 1996 to March 2003. To derive our model we performed a backward stepwise logistic regression analysis including 76 variables in each patient, all of them obtained within 72 h of admission. The primary study outcome was urgent surgery or inhospital mortality. The original prediction rule consisted of three independent predictors of event: one clinical (heart failure); one echocardiographic (periannular complications) and one microbiological (Staphylococcus aureus). After the development of a multivariate analysis, we derived a stratification score for the test cohort using these three variables that have been found to be statistically significant predictors of an event in multivariate analysis. The score was then constructed by a simple arithmetic sum of the number of variables present. Detailed information concerning the derivation process and results was published elsewhere.7
Validation samples
We have validated our model internally in a cohort of 263 patients consecutively admitted from April 2004 to April 2008 at the same centres in which the original prognostic model was derived. Then we tested the accuracy of the model externally in another cohort of 264 patients diagnosed at another university-affiliated tertiary care hospital from 2000 to 2008.
Definition of events
Deaths occurring during hospital stay and urgent surgery, defined as that performed before adequate antibiotic treatment recommended in the guidelines was completed, were regarded as events.9 10 In patients who underwent urgent surgery and died postoperatively, only the surgery was considered. We followed the Duke criteria until 200211 and modified Duke criteria after that for the diagnosis of infective endocarditis.12
Indications for urgent surgery in the derivation and internal validation sample were: (1) left heart failure unresponsive to maximal medical treatment, defined as lack of improvement in signs and symptoms of heart failure despite optimal medical treatment according to the guidelines; (2) septic shock; (3) uncontrolled infection defined as persistent bacteriaemia or fever lasting more than 7 days after starting adequate antibiotic treatment when other causes had been excluded. The echocardiographic demonstration of a periannular complication (abscess, pseudoaneurysm and fistula) was not considered an indication for surgery ‘per se’. Protocols used at the external derivation sample were not exactly the same, as they also considered prosthetic dysfunction as an indication for urgent surgery as well as the isolated presence of an abscess.
Statistics
We compared baselines characteristics between the validation and the original derivation samples using χ2 or Fisher' exact test for categorical variables. Continuous variables were compared with Student' t test or the Mann–Whitney U test when variables were not normal. To assess the model' discriminatory power to predict an event, we compared the area under the receiver operating characteristic curve between the validation and the original derivation samples. For all analysis, a two-tailed p value of less than 0.05 was used to define statistical significance. To assess the accuracy of our model to predict an event, we also compared sensitivity, specificity, positive and negative predictive values and likelihood ratios between derivation and validation samples. The SPSS statistical software package (version 15.0) and EPIDAT (version 3.1) were used.
Results
Outcome events
Derivation sample versus internal validation sample
There were more events in the internal validation sample than in the derivation sample (50% vs 41%, p=0.034). Considering exclusively the subgroup with an event, the proportion of patients urgently operated on was higher and mortality rates was lower in the internal validation cohort (50% vs 63% and 50% vs 37%, respectively, p=0.04; table 1)
Number and type of event in each group for comparison
Derivation sample versus external validation sample
Similar differences were observed in the external validation sample: more events than in the derivation sample (54% vs 41%, p=0.002), a higher proportion of urgent surgery (50% vs 70%) and lower rate of inhospital mortality (50% vs 30%, p=0.001; table 1)
Baseline characteristics of study populations
We compared a total of 76 baseline variables, all of them obtained within 72 h of the admission between the three groups. The most important differences are summarised in tables 2 and 3.
Comparison of baseline characteristics in the derivation and internal validation sample
Comparison of baseline characteristics in the derivation and external validation sample
Comparison of the prognostic model in the derivation and validation samples
Sensitivity, specificity, positive and negative likelihoods ratios and predictive values in the derivation and both validations samples are represented in table 4.
Accuracy of the prognostic model to predict urgent surgery or inhospital mortality among the three groups
The discriminatory power of the model, expressed as the area under the receiver operating characteristic curve, was similar between derivation and both validation samples (internal 0.67 vs 0.68, p=0.79; external 0.67 vs p=0.74, p=0.09).
Application of the risk score in the validation samples
The proportion of patients who developed an event in the validation samples depended on the number and type of prognostic factors shown in figure 1. The risk score was calculated by assigning a value of one when an independent risk variable was present and then categorising patients in the validation cohorts by the number of risk factors present, as shown in figure 1. There was a progressive, significant pattern of increasing event rates as the risk stratification score increased in both validation cohorts (p<0.001 by χ2 for trend).
Rate of events depending on the number and type of variables present.
Discussion
Infective endocarditis is a devastating disease in which the early recognition of predicting factors that identify patients with a high risk of complications is of great interest. Our group recently derived a stratification model based on variables easily obtained within 72 h of admission. Independent risk factors for inhospital mortality or urgent surgery (that performed before antibiotic treatment was completed) were Staphylococcus aureus infection, heart failure and periannular complications. Based on the number and nature of variables present, we derived a model to predict the probability of having an event.7
In this study we have not only prospectively validated the model in our own population, but also in a cohort of patients with a very different profile and not exactly the same indications for surgery, to find out whether the model is generalisable. The differences found between populations might reflect the regional and temporal variations that occur in this disease. Patients in the external validation sample were older and presented to the hospital in a more acute phase of the disease, but have less heart failure and fewer perivalvular complications at admission, which is probably related to the lower proportion of prosthetic valve endocarditis compared with the derivation sample. It is difficult to determine the impact in our results of the different indications for surgery adopted in the external validation cohort, but we hypothesise that the influence is not significant as the proportion of events in patients with abscess or prosthetic dysfunction—‘per se’ indications for surgery only in the external validation sample—was very similar in both validation cohorts. Our results confirmed that our model is applicable and valid in different populations of patients, and therefore it can be used in the prognostic stratification of patients with left-sided endocarditis admitted at institutions with similar characteristics to ours.
Despite the existence of clinical guidelines and consensus documents on infective endocarditis,9 10 13 the management of these patients differs profoundly among institutions. This may be explained, at least in part, by the weak clinical evidence supporting the recommendations (not a single recommendation has an A level of evidence). Unfortunately, as DiSesa14 stated ‘there is still as much art as science in the care of patients with infective endocarditis’. These discrepancies and the different baseline characteristics of the patients make it mandatory to test the appropriateness of the prediction models in cohorts of series from other institutions to determine accurately the usefulness of the models in the real world.
Our model, albeit not perfect, is good enough to be used in clinical practice. From now on, we can count on a robust, simple, rapid and easy bedside assessment tool that identifies patients at high risk of having an adverse event, based on the number of high-risk variables present in each patient, as shown in figure 1. Importantly, the model works well with only three variables that are available very early in the disease span. Other authors have found predictors of events, but all of them included variables obtained late in the clinical course; thus, when the clinicians become aware that the patient bears a high risk, it may be too late for an aggressive therapeutic approach to be undertaken.
We have focused on the most challenging period for patients with infective endocarditis, the early outcome during hospitalisation, which concentrates the higher proportion of mortality. Other authors dealing with endocarditis have attempted to establish the profile of patients with a high probability of having events including a longer follow-up.15–19 Although useful for the management of patients, these studies do not help in the decision-making process in the early phase of the disease. Moreover, their results are quite heterogeneous given the differences in the recruitment of patients (right and left-sided endocarditis,4 16 19–26 complicated27–29 or non-complicated endocarditis), follow-up period (from hospitalisation to 20 years),30 variables included,24 31 nature of the study (retrospective21 24 25 32 or prospective) and the percentage of patients who underwent transeosophageal echocardiography (from 43% to 90%).4 Another unique characteristic of our model is that the statistical model was designed with the purpose of mirroring the clinical approach to a patient, thus variables were analysed in the same manner as when we are confronted with a patient. The first variables included were those obtained within the first day (clinical, analytical, electrocardiographic, radiographic). Then we added the echocardiographic data, which in our centres are available within 48 h, and the last variables introduced in the model were microbiological, which results are available within 72 h for the majority of microorganisms causative of endocarditis.
Once we are able to identify early the high-risk profile patients with infective endocarditis, the following step is to investigate if a more aggressive therapeutic approach for these patients improves their grim prognosis. Therefore, hospitals participated in this study, started in October 2007, the ENDOVAL trial (NCT00624091),33 the first randomised study in infective endocarditis in which two therapeutic strategies are compared for patients with a high-risk profile: emergent surgery (within 48 h after diagnosis is established) and the state-of-the-art strategy recommended in practical guidelines. The results of the study will shed light on the most challenging decision in patients with endocarditis, which is to decide whether and when surgery has to be undertaken in the active phase of the disease.
Conclusions
Our results suggest that the early risk stratification model we derived, based on variables obtained within 72 h ofm admission, is applicable to different populations with left-sided endocarditis. We provide clinicians with a simple bedside assessment tool that identifies patients at high risk of having an adverse event.
References
Footnotes
See Editorial, p 1117
Linked article 218578.
Funding This study was financed in part by the Cooperative Network for Cardiovascular Research (Red Cooperativa de Enfermedades Cardiovasculares, RECAVA) of the Spanish National Institute of Health (Instituto de Salud Carlos III). BA and NFH belong to the Spanish Network for Research in Infectious Diseases (REIPI RD06/0008).
Competing interests None.
Ethics approval This study was conducted with the approval of the ethical committees of all the participing centres.
Provenance and peer review Not commissioned; externally peer reviewed.