The Impact of Valve Surgery on 6-Month Mortality in Left-Sided Infective Endocarditis
Background— The role of valve surgery in left-sided infective endocarditis has not been evaluated in randomized controlled trials. We examined the association between valve surgery and all-cause 6-month mortality among patients with left-sided infective endocarditis.
Methods and Results— A total of 546 consecutive patients with left-sided infective endocarditis were included. To minimize selection bias, propensity score to undergo valve surgery was used to match patients in the surgical and nonsurgical groups. To adjust for survivor bias, we matched the follow-up time so that each patient in the nonsurgical group survived at least as long as the time to surgery in the respective surgically-treated patient. We also used valve surgery as a time-dependent covariate in different Cox models. A total of 129 (23.6%) patients underwent surgery within 30 days of diagnosis. Death occurred in 99 of the 417 patients (23.7%) in the nonsurgical group versus 35 deaths among the 129 patients (27.1%) in the surgical group. Eighteen of 35 (51%) patients in the surgical group died within 7 days of valve surgery. In the subset of 186 cases (93 pairs of surgical versus nonsurgical cases) matched on the logit of their propensity score, diagnosis decade, and follow-up time, no significant association existed between surgery and mortality (adjusted hazard ratio, 1.3; 95% confidence interval, 0.5 to 3.1). With a Cox model that incorporated surgery as a time-dependent covariate, valve surgery was associated with an increase in the 6-month mortality with an adjusted hazard ratio of 1.9 (95% confidence interval, 1.1 to 3.2). Because the proportionality hazard assumption was violated in the time-dependent analysis, we performed a partitioning analysis. After adjustment for early (operative) mortality, surgery was not associated with a survival benefit (adjusted hazard ratio, 0.92; 95% confidence interval, 0.48 to 1.76).
Conclusions— The results of our study suggest that valve surgery in left-sided infective endocarditis is not associated with a survival benefit and could be associated with increased 6-month mortality, even after adjustment for selection and survivor biases as well as confounders. Given the disparity between the results of our study and those of other observational studies, well-designed prospective studies are needed to further evaluate the role of valve surgery in endocarditis management.
Received August 16, 2006; accepted January 16, 2007.
Whether to recommend valve surgery in a patient with complicated left-sided infective endocarditis (IE) can be one of the most difficult management conundrums in the treatment of infectious diseases. Despite the availability of this type of surgery for 4 decades, no randomized controlled trials have been conducted to better clarify its role in the treatment of specific subgroups of patients and its optimal timing to improve outcomes. Therefore, current practice guidelines of the surgical management of complicated left-sided IE are largely based on results of observational studies and expert opinion.1
Clinical Perspective p 1728
To better evaluate the impact of valve surgery on 6-month mortality in adult patients with complicated left-sided native valve endocarditis, investigators recently used a propensity score analysis to adjust for the confounding effect of treatment selection bias.2 More specifically, they developed a propensity score that quantitatively predicted the likelihood that surgery would be recommended for an individual patient. Next, the investigators matched surgical patients with nonsurgical patients by propensity scores and analyzed outcome (6-month mortality). Their results indicated that surgical intervention was beneficial in patients with underlying moderate-to-severe congestive heart failure.
This was the first time that a propensity analysis had been used to examine management issues in IE. Thus, the attributes of this methodology were lauded in an accompanying editorial3 with expectations for future use of this statistical tool to examine heretofore difficult questions in cardiovascular infectious diseases. The results of this nascent investigation were astutely recognized as both “surprising and provocative.” The question was posed, “Is it possible that the presumed benefits of surgery are overestimated for patients who do not have significant heart failure?”3 More recently, 2 studies4,5 from the International Collaboration on IE Merged Database reported on the association of valve surgery and in-hospital mortality in patients with native and prosthetic valve IE (PVE), determined with the same methodology. These studies did not find a uniform and statistically significant improvement in survival in patients who underwent surgery. However, their analysis was limited by an incomplete adjustment for confounders and missing echocardiographic data. Moreover, the 3 previous studies did not account for the possibility of survivor treatment selection bias,6 which can cause conflicting results, as demonstrated, for example, in recent studies of the relationship of blood transfusion and clinical outcomes among patients with acute coronary syndromes.7,8
Therefore, we conducted a cohort study of patients with left-sided IE to examine the association of valve surgery and 6-month mortality. We adjusted for treatment selection and survivor biases with propensity score and time-dependent covariate analyses.
Consecutive patients who were 18 years of age or older and had been diagnosed and treated for left-sided IE at the Mayo Clinic in Rochester, Minnesota, between 1980 and 1998 were eligible for inclusion in the study. Patients were excluded if they refused to consent to medical record review for research purposes or if patients left the hospital before a complete diagnostic work-up or treatment plan was outlined. Patients’ medical records were reviewed to confirm the diagnosis according to the modified Duke criteria.9 The institutional review board at Mayo Clinic approved the study and waived the requirement for informed consent.
Case definitions of IE have changed over time and include the widely used Beth Israel,10 Duke,11 and modified Duke criteria.9 Cases were identified in the current investigation with the modified Duke criteria.9 We screened all IE cases that were defined before the introduction of Duke criteria in 1994 and applied the new case definition retrospectively.
Functional class at presentation was defined according to the New York Heart Association classification from 1 to 4. The Charlson comorbidity score contains 19 categories of comorbidity that are primarily defined with diagnostic codes.12–14 Charlson score has been previously validated as an independent predictor of 6-month mortality in patients with left-sided endocarditis.15
Stroke was defined according to the World Health Organization definition16: a neurological deficit that lasts >24 hours and is of presumed vascular origin. Valve surgery included replacement or repair of the affected valve. To assess the presence of vegetation and severity of valvular dysfunction, we used data from transesophageal echocardiography preferentially, and if data were not available we relied on transthoracic echocardiography.
The primary outcome was all-cause 6-month mortality after date of IE diagnosis. We chose all-cause mortality because it is a robust outcome17 and allowed us to compare our results with published studies.2 Ascertainment of death at 6 months was obtained from medical records; this process was supplemented by the recording of all obituaries and notices of deaths by the Mayo Clinic registration office and with the use of the National Death Index, which has been previously validated in the determination of the date of death.18
We examined the association between valve surgery and 6-month mortality after adjustment for important biases and confounders that may have affected previous studies. First, in a nonrandomized observational study, investigators have no control over the treatment assignment, and therefore direct comparison of outcomes may mislead investigators because of selection bias or “confounding by indication.”19,20 Selection bias may be reduced with propensity score analysis.21,22 The propensity score is the conditional probability of valve surgery given an individual vector of measured covariates. Matching on this score can yield a quasi-randomized trial. Another approach to minimize selection bias is to use this score in the regression adjustment of the final estimate of the treatment effect.21,22
Second, longer survival may increase a patient’s chance to use treatment in an observational study. Patients who die sooner have less time to be selected for treatment and thus by default are more likely to remain untreated. If this is not considered in the data analysis, the estimate of the treatment effect will be affected by “survivor treatment selection bias.”6 Survivor bias can be eliminated by adjustment for the time at which surgery is performed. This can be accomplished by a match on the follow-up time such that the patient in the nonsurgical group survived at least as long as the time to surgery in the surgically treated patient. Alternatively, the time of surgery can be considered in proportional hazards analyses with a time-dependent covariate analysis. A time-dependent covariate is a variable whose value is allowed to change with the time component in the model.23
Propensity Score Calculation
Baseline variables, complications that develop during hospitalization (eg, stroke or other embolic events), and variables that occur after discharge such as IE relapse, which were associated with valve surgery, were used to derive a model to predict a propensity score for each patient. These variables were derived from a stepwise logistic regression model where surgery is the outcome.21 Clinically relevant (eg, Staphylococcus aureus, PVE, acute valvular dysfunction, severe chronic heart failure) and statistically significant variables in this logistic regression analysis were added to derive a full nonparsimonious model. Clinically plausible interactions were also included in this analysis. Each patient had a predicted propensity score based on his variables, which reflected his probability to undergo surgery.
Modeling of Outcomes
We used 2 modeling methods to examine the association between surgery and 6-month mortality.
Method 1 (Matched Cohort)
Patients who underwent surgery were matched to patients who did not undergo surgery in a 1-to-1 fashion24 on the basis of the following variables: (1) logit of propensity score (log [p/(1−p)]) within ±0.25 SD(logit)21; (2) diagnosis decade (1980 to 1989, 1990 to 1998); and (3) follow-up time such that the patient in the nonsurgical group survived at least as long as the time to surgery in the surgically treated patient. Cox proportional hazards regression25,26 was then used to determine the association between surgery and 6-month mortality. The analysis accounted for the matching between surgical cases and nonsurgical referent cases with the SAS procedure PHREG (SAS Institute, Cary, NC) with a STRATA statement to account for the matching.
Method 2 (Surgery as a Time-Dependent Covariate)
The total cohort was considered for this analysis. A time-dependent covariate data structure27 was created and analyzed with Cox proportional hazards regression26 with the counting process style of input in PROC PHREG (SAS Institute). Surgery was used as a time-dependent covariate through 2 different approaches: first as a time-dependent covariate without lag (ie, surgical patients were included in the nonsurgical group until the date of surgery and in the surgical group on the date of surgery, and second as a time-lagged covariate with a 3-day lag23 (ie, surgical patients were included in the nonsurgical group until 3 days before the date of surgery and in the surgical group after that point). The 3-day lag was chosen on the basis of the clinical impression that, on average, a 3-day lag exists between the decision to take the patient for valve surgery and the actual date of surgery.
All models were adjusted for prognostic variables known to be predictive of 6-month mortality and the logit of propensity scores from the model described above. The robust sandwich estimator of variance (COVSANDWICH option, SAS Institute) was used to account for patients with multiple episodes included in the analysis. The proportional hazards assumption was assessed on the basis of a test of Schoenfeld residuals with the cox.zph procedure in S-plus (version 6.2 for Windows, Insightful Corp., Seattle, Wash).
To compare our results with those by Vikram et al,2 we restricted our cohort to patients with complicated left-sided native valve IE so that we could repeat the analysis in the second method (surgery was used as a time-dependent covariate with a 3-day lag). We defined “complicated IE” in the presence of any of the following: definite vegetation, cardiac abscess, stroke or other embolic events, chronic heart failure (New York Heart Association class 3 or 4).
To assess the consistency of the observed effect estimates, we performed a priori planned subgroup analyses with the Cox proportional hazards regression models with interaction terms.26 Interaction models were assessed with the second method (surgery was used as a time-dependent covariate with a 3-day lag). Hazard ratios (HRs) for the effect of surgery were estimated within each subgroup with the parameter, variance, and covariance estimates from the interaction model. Two-sided P value <0.05 was considered to be statistically significant for all analyses. Analyses were computed with SAS software, version 8.2 (SAS Institute Inc, Cary, NC).
The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agree to the manuscript as written.
The cohort included 546 IE cases diagnosed between 1980 and 1998; of these, 512 cases (93.77%) met the “definite IE” definition by the modified Duke criteria, and 129 cases (23.6%) had valve surgery in the first 30 days after IE diagnosis. Table 1⇓ displays the characteristics of the cohort.
Propensity Score Calculation
The propensity model included age, sex, diagnosis year, Charlson comorbidity score, mitral valve involvement, New York Heart Association class, PVE, hemoglobin level, valvular dysfunction status, presence of vegetation, presence of large vessel emboli, Janeway lesions, relapse, cardiac abscess, stroke, viridans group streptococci, coagulase negative staphylococci, fungi, S aureus, Gram-negative bacilli, and interaction terms of PVE with S aureus and coagulase negative staphylococci. Table Ia in the online-only Data Supplement summarizes the odds ratios associated with each of the variables in the model. The C-statistic for this model was 0.856, which indicated a strong ability to differentiate between patients who did and did not undergo valve surgery.
We were able to match 93 of 129 patients in the surgical group to 93 of 417 patients in the nonsurgical group. A comparison of different characteristics of the propensity-matched cohort is displayed in Table IIa of the online-only Data Supplement. After matching, no significant differences existed between the 2 groups.
Surgery and 6-Month Mortality
Of the 417 patients in the nonsurgical group, 99 (23.7%) died, versus 35 (27.1%) who died in the 129 patients in the surgical group. Eighteen of 35 patients (51%) in the surgical group died within 7 days after valve surgery. Table 2 summarizes the results of the different Cox models that examined the association between valve surgery and mortality.
Of the 93 patients in the propensity-matched subset, 18 (19.4%) in the nonsurgical group died, versus 27 (29%) who died in the 93 patients in the surgical group. After adjustment for the logit of propensity score and potential confounding variables, valve surgery was associated with an HR of 1.3 (95% confidence interval [CI], 0.5 to 3.1; P=0.56].
HRs were also calculated with the entire cohort of patients between 1980 and 1998. When surgery was treated as a time-dependent covariate, valve surgery was associated with an increase in the 6-month mortality with an adjusted HR of 1.9 (95% CI, 1.1 to 3.2; P=0.03). When a 3-day lag was considered, the adjusted HR for mortality was 1.5 (95% CI, 0.9 to 2.6; P=0.11). When the analysis was restricted to “complicated” left-sided native valve IE and a 3-day lag was considered, adjusted HR for mortality was 1.3 (95% CI, 0.6 to 2.6; P=0.50).
Clinically important variables were tested for possible interactions with the association of valve surgery and mortality. Results are displayed in Table 3. A statistically significant interaction existed between the effect of surgery and the presence or absence of vegetation and the involvement or noninvolvement of the mitral valve.
Proportional Hazard Assumption
The proportional hazard assumption for the surgery variable was violated for both the time-dependent covariate analysis without lag (Schoenfeld residuals test,26 P<0.05), and the 3-day lag analysis (Schoenfeld residuals test, P=0.06). The plot of residuals over time (Figure 1) for the time-dependent covariate analysis without lag shows the nature of the nonproportionality.
Examination of the distribution of death after surgery revealed an early high risk of death within 7 days of surgery, which resulted in a nonconstant coefficient for surgery over time and thus nonproportional hazards. To deal with nonproportional hazards and to account for this early surgical risk, we partitioned the surgical variable into 2 components with step functions. Two time-dependent surgery variables were created. The first variable (X1) indicated whether the patient had surgery within the last 7 days and captured the short-term effect (ie, operative mortality), and the second variable (X2) captured the longer-term surgical effect after adjustment for the short-term effect. A Cox model that was adjusted for the same variables as the previous models described above and contained X1 and X2 revealed an adjusted HR of 6.21 (95% CI, 2.72 to 14.18) and 0.92 (95% CI, 0.48 to 1.76), for X1 and X2, respectively. Therefore, even after adjustment for early (operative) mortality (ie, increased risk during the first week after surgery), surgery was not associated with a survival benefit (adjusted HR, 0.92; 95% CI, 0.48 to 1.76).
We found that valve surgery in our patient population with left-sided IE has no survival benefit and could be associated with increased 6-month mortality. When included as a time-dependent covariate with no lag or with a 3-day lag in the survival model, surgery was associated with both a significantly higher risk of death and a trend toward higher risk of death, respectively. In the matched cohort analyses, no survival benefit existed with surgery with the point estimate consistent with harm; the number needed to treat with valve surgery to harm 1 patient was 20 (95% CI, benefit 11 to ∞ to harm 3).27 This increased risk of death appeared mainly in the early postoperative period and may be attributed to operative mortality. Nevertheless, even after adjustment for early (operative) mortality (ie, if patients survive the first week after surgery), surgery was not associated with a survival benefit.
Whether a subgroup of patients exists who may benefit from surgery remains an important question that could not be answered by our study and requires further investigation. The results of the subgroup analyses are limited as post-hoc analyses (although planned a priori) and by the small number of events in specific subgroups that make it difficult to find significant associations. The HR point estimates are suggestive of benefit from surgery in patients with definite vegetation, mitral valve involvement, and cardiac abscess because the estimates are <1, but the 95% CI upper limit suggests the possibility of significant harm.
Our results run counter to conventional clinical thinking about the presumed benefit of valve surgery in the treatment of IE. Biological plausibility exists for the benefit of surgery in IE. Removal of the focus of infection and repair of valvular dysfunction is intuitively a surrogate of improved survival. Nevertheless, the results of surgery depend on many factors that include the general preoperative condition of the patient, antibiotic treatment, timing of surgery, perioperative management, surgical techniques, and postoperative management.28 The reason for this paradox (surgery associated with worse outcome) is unclear. Because of the recognized surgical expertise provided by our institution, the complexity of cardiovascular disease, and the multiple comorbidities that characterize the referral population, referral bias must be considered.
Comparison to Other Studies
Recently published studies that used propensity analysis to examine the impact of surgical intervention have provided varying results.2,4,5,29 Although the studied cohorts were chosen across different time periods, our cohort is relatively contemporary (1980–1998) and comparable to the cohort by Vikram et al2 (1990–2000) and by Mourvillier et al29 (1993–2000).
In the first rigorously conducted study, Vikram et al2 analyzed data from a multicenter cohort and found that surgery, used as a time-fixed covariate, was associated with a lower 6-month mortality among 218 patients with complicated left-sided native valve endocarditis who were matched by propensity scores, with an adjusted HR of 0.4, (95% CI, 0.18 to 0.91). The authors acknowledged, however, that their definition of baseline as either the date of surgery or the date of the decision not to operate has methodological disadvantages because it prevented an analysis of the potential impact of timing of surgery on the observed association. Moreover, the date of the decision not to operate is theoretical and difficult to ascertain as a result of the evolving nature of IE, the development of surgical indications over time, and the difficulty in defining a specific day that this decision is made. Finally, the investigators adjusted for Charlson comorbidity score as a dichotomous variable (<2 or ≥2). The Charlson comorbidity score contains 19 categories of comorbidity.12–14 The overall comorbidity score reflects the cumulative increased likelihood of 1-year mortality: the higher the score, the more severe the burden of comorbidity. Therefore, categorization of this score into 2 groups (<2 or ≥2) could violate the comparability of the surgical and nonsurgical groups and potentially bias the results.
In contrast to the findings of the Vikram et al investigation,2 data from 2 medical intensive care units demonstrated no significant difference in in-hospital mortality between surgical and nonsurgical groups.29 Mourvillier et al developed a propensity score model that contained new valve regurgitation on echocardiography, mechanical ventilation, diabetes mellitus, the absence of neurological complications, and community-acquired native valve endocarditis.29 Of 228 patients, 27 surgical patients were successfully matched with 27 nonsurgical patients. Surgical patients were not at a lower risk of hospital death (adjusted odds ratio, 0.96; P=0.95).
In 2 other recent studies,4,5 investigators from the International Collaboration on Endocarditis Merged Database, a prospective cohort from 7 sites in 5 countries, reported on the association of valve surgery and in-hospital mortality in patients with native and PVE. In their analysis of 1516 patients with native valve endocarditis,4 surgery was associated with a survival benefit only in the subgroup of patients in the highest propensity scores quintiles. This subgroup consisted of 249 patients who underwent surgery versus only 50 patients who did not, with respective mortality rates of 11.2% versus 38%. However, this analysis, although restricted to the highest propensity quintile, was not adjusted for potential confounders.
In their other analysis of 367 patients with PVE,5 they found a statistically nonsignificant survival benefit of surgery in the subset of 136 patients matched by their propensity scores, with mortality rates of 22.1% and 32.4% in patients who did and did not undergo surgery, respectively (P=0.18). Similar findings were noted in the subset with highest propensity scores with an adjusted HR of 0.61 (95% CI, 0.26 to 1.44). Although their analysis is distinguished by the relatively large sample size, it is limited by the availability of important collected data. In fact, the database is merged from different independent databases that contain a heterogeneous set of variables with different definitions. For example, it does not contain data on the timing of surgery, all important patient comorbidities, or echocardiographic information that are well known to affect mortality. Therefore, a comparison of different groups of patients in this cohort without complete adjustment for an unequal distribution of these important variables could lead to biased estimates.
Strengths and Limitations
The present study has several important strengths. First, we used rigorous case definitions and derived a propensity score model from most known predictors of surgery. Second, we adjusted for survivor bias. The analysis of all 3 previously published studies2,4,5 did not consider survivor bias. If this factor is not evaluated in the analysis, the estimate of the treatment effect will be biased toward a beneficial effect.6 A recent systematic review showed that survivor bias is surprisingly common in studies published in top medical journals. In more than half of the identified investigations, correction of this bias would have changed the respective study’s conclusions.30 Third, we adjusted for stroke as a time-dependent covariate to take into account the timing of stroke with respect to surgery, which has not been considered in previous studies. Finally, we performed several analyses that supported the consistency and robustness of our findings.
Limitations to our study exist. Although we used several rigorous statistical methods to adjust for the severity of disease, the causative microorganisms, and host-related factors, there may still be unmeasured confounders that could account for increased surgical mortality. Moreover, despite rigorous analyses, the effect of surgery on survival may still be in part confounded by indication or other selection biases that can occur in any observational study.20,31 Finally, referral bias can limit the applicability of our findings.
The results of our study suggest that valve surgery in left-sided IE has no survival benefit and could be associated with increased 6-month mortality. This mortality risk persisted despite minimization of selection and survivor biases and confounding. Even if patients were to survive the early postoperative period, surgery conferred no survival benefit. Because our study was not randomized, its findings should not be considered as evidence to change practice; rather, our study provides further evidence that scrutiny is warranted when management decisions in complicated left-sided IE are made. Given the disparity in results between our study and other observational studies, well-designed prospective investigations that address the methodological issues that we have highlighted are needed to further evaluate the role of valve surgery in endocarditis management.
This work is part of a thesis that was submitted by the first author to the Faculty of Biomedical Sciences, Mayo Clinic College of Medicine, and was approved in partial fulfillment of the requirements for the Master’s degree in Sciences–Clinical Research.
Sources of Funding
The present study was supported by grants from the Infectious Diseases Division Small Grants Program and the ENHANCE Award from the Department of Medicine, Mayo Clinic.
Baddour LM, Wilson WR, Bayer AS, Fowler VG Jr, Bolger AF, Levison ME, Ferrieri P, Gerber MA, Tani LY, Gewitz MH, Tong DC, Steckelberg JM, Baltimore RS, Shulman ST, Burns JC, Falace DA, Newburger JW, Pallasch TJ, Takahashi M, Taubert KA; Committee on Rheumatic Fever, Endocarditis, and Kawasaki Disease; Council on Cardiovascular Disease in the Young; Councils on Clinical Cardiology, Stroke, and Cardiovascular Surgery and Anesthesia; American Heart Association; Infectious Diseases Society of America. Infective endocarditis: diagnosis, antimicrobial therapy, and management of complications. Circulation. 2005; 111: 3167–3184.
Cabell CH, Abrutyn E, Fowler VG Jr, Hoen B, Miro JM, Corey GR, Olaison L, Pappas P, Anstrom KJ, Stafford JA, Eykyn S, Habib G, Mestres CA, Wang A; International Collaboration on Endocarditis Merged Database (ICE-MD) Study Group Investigators. Use of surgery in patients with native valve infective endocarditis: results from the International Collaboration on Endocarditis Merged Database. Am Heart J. 2005; 150: 1092–1098.
Wang A, Pappas P, Anstrom KJ, Abrutyn E, Fowler VG Jr, Hoen B, Miro JM, Corey GR, Olaison L, Stafford JA, Mestres CA, Cabell CH; International Collaboration on Endocarditis Investigators. The use and effect of surgical therapy for prosthetic valve infective endocarditis: a propensity analysis of a multicenter, international cohort. Am Heart J. 2005; 150: 1086–1091.
Li JS, Sexton DJ, Mick N, Nettles R, Fowler VG Jr, Ryan T, Bashore T, Corey GR. Proposed modifications to the Duke criteria for the diagnosis of infective endocarditis. Clin Infect Dis. 2000; 30: 633–638.
Von Reyn CF, Levy BS, Arbeit RD, Friedland G, Crumpacker CS. Infective endocarditis: an analysis based on strict case definitions. Ann Intern Med Medicine. 1981; 94: 505–518.
Newman TB, Brown AN. Use of commercial record linkage software and vital statistics to identify patient deaths. J Am Med Inform Assoc. 1997; 4: 233–237.
Hak E, Verheij TJ, Grobbee DE, Nichol KL, Hoes AW. Confounding by indication in non-experimental evaluation of vaccine effectiveness: the example of prevention of influenza complications. J Epidemiol Community Health. 2002; 56: 951–955.
Bergstralh EJ, Kosanke JL. Computerized Matching of Cases to Controls. Technical Report Series No. 56. Rochester, Minn: Section of Biostatistics, Mayo Clinic, 1995.
Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York: Springer-Verlag; 2000.
Altman DG, Andersen PK. Calculating the number needed to treat for trials where the outcome is time to an event. BMJ. 1999; 319: 1492–1495.
Whether to recommend valve surgery in a patient with left-sided infective endocarditis (IE) can be one of the most difficult management conundrums in the treatment of infectious diseases. Because of the uncommon occurrence of IE, no randomized controlled trials have been conducted to better clarify its role in IE management. Recently, investigators have evaluated the role of surgery among large prospective and retrospective cohorts of IE patients. Because of the lack of randomized controlled surgical treatment trials, these investigators used a propensity score methodology to decrease the impact of selection bias that typically accompanies observational studies. They reported improved survival among IE patients who undergo valve surgery. Nevertheless, their analyses did not consider the timing of surgery and the potential impact of survivor bias. The present study presents data from a large cohort of patients with left-sided IE. The authors examined the association of valve surgery and 6-month mortality after adjustment for treatment selection and survivor biases. They found that valve surgery has no survival benefit and could be associated with increased 6-month mortality. Because the study was not randomized, its findings should not be considered as evidence to change practice; rather, it provides further evidence that scrutiny is warranted when management decisions for IE are made. Given the disparity between the results of this study and those of other observational studies, well-designed prospective investigations that address the methodological limitations are needed to further evaluate the role of valve surgery in IE management and to define the group of patients that would benefit from this intervention.
The online-only Data Supplement, consisting of tables, is available with this article at http://circ.ahajournals.org/cgi/content/ full/CIRCULATIONAHA.106.658831/DC1.