Short-Course Therapy of Catheter-related Staphylococcus aureus Bacteremia

Methods

Using MEDLINE (1966 to 1991) and bibliographic review of relevant articles and textbooks, we searched the literature for studies which offered data specifically regarding catheter-related S. aureus bacteremia. Studies that dealt exclusively with immunocompromised patients were excluded. Eleven studies were identified that reported late complication rates after short-course ( 2 weeks) therapy [3-7, 9, 19, 22-25]. For the purposes of this study, we defined a late complication as an infectious complication occurring after therapy has been completed. This classification includes not only cases with documented relapse of S. aureus bacteremia or metastatic S. aureus infection (for example, epidural abscess) but also complications clinically attributable to the initial infection even in the absence of positive culture data (for example, development of culture-negative endocarditis several weeks after completion of short-course therapy for S. aureus bacteremia). The methods section of each study was carefully reviewed for evidence of protection against four important types of bias. In addition, we extracted data from any study, regardless of treatment duration, reporting complication or mortality rates or both for catheter-related S. aureus bacteremia.

In the sections that follow, we provide the details of the criteria used to assess protection against bias and their application to the individual studies. Note that these criteria were applied specifically to assess the validity of the estimate of rate of late complications after short-course therapy for uncomplicated catheter-related S. aureus bacteremia and not to other hypotheses addressed by these studies.

Protection against Diagnostic Misclassification Bias

It can be difficult to determine which bloodstream infections are caused by an infected catheter. Unless careful criteria are followed, there is significant risk for misclassification when identifying those bacteremias that are catheter-related. Diagnostic misclassification bias occurs when misclassified cases have an outcome systematically different from those that were correctly classified. We defined a study as having provided adequate protection against diagnostic misclassification bias if both of the following were true: 1) each patient was examined to rule out infectious complications including endocarditis before or during antibiotic therapy; and 2) a positive blood culture for S. aureus with the same antibiogram demonstrated for isolates from either a quantitative [26, 27] or semiquantitative [28] catheter segment culture or from a culture of pus at the catheter site was required to define each case of catheter-related S. aureus bacteremia.

Protection against Sampling Bias

Sampling bias exists if patients included in a study represent a subset of those originally eligible and the subset happens to be systematically different from the larger population from which it is drawn. We defined a study as having provided reasonable evidence of protection against sampling bias if one or more of the following characteristics were present: 1) all cases identified in the reporting hospitals during the study period were included; 2) if all were not included, detailed justification of the reasons for exclusion were provided; and 3) if all were not included, comparison data were provided to demonstrate the absence of systematic differences between those included and those excluded. We required that evidence for protection against bias be present not only in identifying cases of S. aureus bacteremia, but also in choosing those cases that were specifically catheter related.

Protection against Treatment Allocation Bias

Treatment allocation bias occurs when those selected for the therapy under study are systematically different from those to whom they are compared. Controlled studies were defined as having provided evidence for protection against treatment allocation bias if at least one of the following were true: 1) treatment was randomly allocated; or 2) comparison data regarding potential confounding variables were presented to demonstrate similarity between those who received and those who did not receive short-course therapy.

Uncontrolled studies were defined as having provided evidence of protection against treatment allocation bias if at least one of the following was true: 1) all eligible patients with uncomplicated catheter-related S. aureus bacteremia received short-course therapy; 2) a random selection of all eligible patients with uncomplicated catheter-related S. aureus bacteremia received short-course therapy; or 3) comparison data were provided to demonstrate similarity between those selected and those not selected to receive short-course therapy.

Protection against Surveillance Bias

In studying the safety of short-course therapy, surveillance bias occurs when patients are not followed long enough to detect late complications. We defined studies as having adequate protection against surveillance bias if the participants were clinically followed for at least 7 weeks after completion of treatment because two of seven (29%) published late complications did not occur until 7 weeks after completing short-course therapy (Table 1).

Statistical Precision

The statistical precision for each of the studies was assessed. For each study a 95% confidence interval was calculated for the proportion of cases of catheter-related S. aureus bacteremia that developed late complications after the patient received short-course therapy. Normal approximation methods were used where applicable, otherwise exact binomial methods were used [29, 30].

Results

Eleven studies were identified including 132 patients treated with short-course therapy for uncomplicated catheter-related S. aureus bacteremia. The clinical characteristics of the patients and types of catheters were not described in most of these studies (Table 2).

Protection against Diagnostic Misclassification Bias

No study fulfilled both criteria for protecting against diagnostic misclassification bias. Although semi-quantitative techniques were used in three studies for diagnosing some cases [6, 9, 23], no study required such data to document all cases. Nine of the 11 studies [3-7, 9, 22-24] state specifically that patients were examined for evidence of infectious complications before or during antibiotic therapy.

Protection against Sampling Bias

Table 3 shows which studies fulfilled the criteria for protection against sampling bias. One study [3] retrospectively reviewed 22 cases of catheter-related S. aureus bacteremia selected from 293 cases of S. aureus bacteremia. It was not clearly stated that all initially uncomplicated cases of catheter-related S. aureus bacteremia were included. Another study excluded an unknown number of patients with no identifiable primary focus [5]. A third study excluded one half of cases of catheter-related S. aureus bacteremia without providing detailed justification or comparison data [7]. One group of investigators [22] excluded almost 30% of S. aureus bacteremia cases because no serologic data were available; although this was appropriate with regard to another question addressed in their study, this exclusion may have biased the outcome observed after short-course therapy. In addition, their study evaluated relapse after short-course therapy in only that portion of their patients who, in retrospect, appeared to have benign staphylococcal bacteremia. This implies that patients with poor outcomes were not included. Another group of investigators only included patients seen in consultation by the Infectious Disease Division of their medical center. This may have resulted in biased selection of the most severe or complicated cases [4].

Protection against Treatment Allocation Bias

In only three of the studies was there adequate protection against treatment allocation bias: those of Rahal and colleagues [24], Bayer and colleagues [5], and Graham and colleagues [25]. The study by Rahal provided the highest quality protection through random allocation of therapy. Bayer and associates [5] and Graham and coworkers [25], although using uncontrolled designs, treated all recognized patients with uncomplicated catheter-related S. aureus bacteremia with short-course therapy. In seven studies [3, 4, 6, 7, 9, 19, 23], only a subset of all recognized cases of catheter-related S. aureus bacteremia received short-course therapy and the treatment choices were made not by the investigator but rather the primary physician. None of these seven studies provided comparison data for those patients not chosen for short-course therapy. One group of investigators [22], by reporting late complication rates only for a subset of their patients who were retrospectively determined to have benign catheter-related S. aureus bacteremia, made it difficult to interpret the effectiveness of short-course therapy using their data.

Protection against Surveillance Bias

Only four studies provided evidence that clinical follow-up was routinely maintained for at least 7 weeks after completion of therapy [3, 7, 9, 24].

Statistical Precision

Ninety-five percent confidence intervals for each study reviewed are shown in Figure 1.

Summary

Table 4 summarizes the findings of our methodologic evaluation and gives the late complication rates reported by each study. The studies are listed in decreasing order of quality of protection against bias where highest priority was given to studies protecting against allocation bias. The order was determined by counting the number of criteria that were met by each study. In the case of a tie, priority was assessed in the following order: first, to studies that protected against allocation bias, followed by prospective studies, and finally, studies that protected against surveillance bias. Table 5 gives published morbidity and mortality data for catheter-related S. aureus bacteremia regardless of which therapeutic regimen was used.

Discussion

The hypothesis that short-course therapy may be adequate in treating catheter-related S. aureus bacteremia derives from the suggestion by several authors that this is an infection with a benign prognosis and a low risk for complications [3, 6, 18, 32, 40, 45]. The findings of this review do not support these suggestions. The available published data demonstrate that almost one of four patients will develop complications and one of seven will die from this infection. Although data are sparse, it seems that relapses after therapy of catheter-related S. aureus bacteremia are at least as dangerous as the initial infection. It therefore seems important to choose a treatment with a low probability of late complications.

A simple approach to determining the late complication rate for this regimen would be to pool the results of all available studies regardless of methodologic approach. The combined late complication rate for all 11 studies of short-course therapy for catheter-related S. aureus bacteremia is 6.1% (95% CI, 2.0% to 10.2%). If the true rate of late complications with short-course therapy approaches the upper confidence limit of this estimate, it would probably be regarded by most clinicians as unacceptably high given the severity of reported late complications (see Table 1). Thus, even without regard to the methodologic quality of individual studies, the safety of short-course therapy for this disease must be considered questionable at best. When careful methodologic appraisal is applied to these data, it is not clear that simple statistical pooling would be valid because of the extreme variability in quality of study design among the studies included in this review [that is, pooling biased results would not necessarily yield a closer approximation of the truth]. If the validity of pooling cannot be accepted, then the true rate of late complications after short-course therapy for catheter-related S. aureus bacteremia becomes even less clear than would be suggested by the confidence interval on the pooled estimate.

When investigating new therapeutic regimens, the randomized, controlled trial is universally accepted as the ideal study design. Some argue that this is the only valid method [46], whereas others take a less adamant stance on the issue [47]. In the setting of catheter-related S. aureus bacteremia, an uncontrolled study may be justified if one chooses the participants in an unbiased manner and if one can assume that the expected late complication rate for catheter-related S. aureus bacteremia after conventional treatment durations is very near zero. Although data are limited, this assumption can be supported by examining the usual clinical outcome of S. aureus endocarditis. It is believed that most late complications after short-course therapy for catheter-related S. aureus bacteremia result from relapses due to inadequately treated occult endocarditis. On the basis of these assumptions, an estimate of the late complication rate after treating catheter-related S. aureus bacteremia with conventional regimens can be calculated by multiplying the relapse rate for S. aureus endocarditis by various estimates of the prevalence of occult endocarditis among patients presenting with catheter-related S. aureus bacteremia (Table 6). Even for the most extreme published estimate of the prevalence of endocarditis in catheter-related S. aureus bacteremia, the relapse rate after conventional 4- to 6-week antibiotic regimens for catheter-related S. aureus bacteremia is expected to be less than 1%. It therefore seems reasonable for an investigator to use an uncontrolled study design because the relapse rate after conventional therapy is predictably close to zero. Nevertheless, an uncontrolled study is susceptible to bias in selecting the patients who are to receive experimental therapy. Only two [5, 25] of the 10 uncontrolled studies provided evidence of protection against this form of bias.

Rahal and colleagues [24] performed the only randomized trial that addressed the late complication rate after short-course therapy for catheter-related S. aureus bacteremia. No other studies provided the quality of protection against bias that is present in this study. Rahal's study was small, and the results showed no significant difference in those who completed short-course therapy as compared to 4 weeks of antibiotic therapy (two of seven patients in the short-course group had a relapse compared with none of 4 in the 4-week group; P = 0.5). Although the statistical precision is low, it is of interest that the late complication rate found by Rahal and associates [24] is considerably higher than that observed in any other study. This may be due to chance, but it is also possible that the differences in methodologic quality might account for the difference in rate seen when comparing the study by Rahal and coworkers [24] with other studies.

Several criticisms of this methodologic evaluation might be raised. One might challenge the validity of the criteria used to evaluate the studies. The rationale for these criteria, which are based on fundamental epidemiologic principles, deserves further explanation.

Because catheter-related bacteremia can be difficult to diagnose, the risk for misclassification is high. To minimize this risk, objective definitions for catheter-related bacteremia should ideally be used. Maki and coworkers [28] showed the need for at least semi-quantitative catheter-segment cultures for documenting catheter-related blood stream infections. Other investigators have confirmed this observation, and quantitative or semi-quantitative techniques of culturing catheter segments have become widely accepted as more accurate and objective methods of identifying catheter-related bloodstream infections [26, 51-54]. Failure to routinely use quantitative or semi-quantitative culture techniques to define cases of catheter-related S. aureus bacteremia may be a potentially significant source of bias due to misclassification because only about 30% of cases of catheter-related bacteremia have physical signs of inflammation at the catheter insertion site [55]. Without quantitative or semi-quantitative culture techniques, it is possible that cases of catheter-related S. aureus bacteremia without physical signs implicating the catheter as the source of the infection may have been systematically excluded. Because some investigators may have included bacteremias in patients with indwelling catheters unless another source could be implicated clinically, cases of non-catheter-related bacteremia may have been included. In either case, bias would result if these misclassified cases have an outcome systematically different from those that were classified correctly.

Sampling bias exists if participants included in a study represent a subset of those originally eligible and the subgroup happens to be systematically different from the larger population from which it is drawn. Such biased sampling may invalidate attempts to generalize the findings to the reference population.

Perhaps the most important criteria applied to these studies were those assessing protection against treatment allocation bias. In all therapeutic studies, the outcome of patients treated with experimental therapy is compared to that of those treated with a standard regimen. In the most valid experimental design, the study includes both an experimental and control group to which treatment is randomly allocated, and their outcomes are compared. If no control group is included, the outcome of the study group is usually compared with the outcome expected based on historical data with standard regimens. In either design, the allocation of treatment should assure that those selected for experimental therapy are comparable to those to whom they are compared. For controlled studies this means the assignment to treatment group must be unbiased. In uncontrolled studies those who receive the experimental therapy should be an unbiased selection from all cases of the disease being studied (for example, uncomplicated catheter-related S. aureus bacteremia). If the allocation of treatment is biased, then the results may be invalid.

Because S. aureus bacteremia and its severe complications are not believed to exist as a subclinical disease, we accepted simple clinical follow-up as adequate surveillance for late complications. Although the optimal duration of follow-up necessary to detect late complications is not known, the data from Table 1 can be used as a guideline. Outcome events may have been systematically missed in studies that did not follow patients for at least 7 weeks after completing therapy.

These criteria should not be considered excessively strict since they represent minimum safeguards necessary to protect against bias. Indeed, most investigators use criteria more stringent than these when designing studies of therapeutic efficacy such as double-blind random allocation to experimental and control regimens.

Some of the studies in this analysis include patients treated for fewer than 10 days. This raises the concern that the reported late complication rates might have been biased toward an overestimate since some patients may not have received adequate therapy. To address this issue, we recalculated the pooled estimate excluding any patients in three studies [5, 7, 9] known to be treated less than 10 days and patients in three studies [3, 6, 23] for whom the precise duration of therapy could not be determined; after these exclusions, the pooled late-complication rate for remaining patients was 8.2% (CI, 2.7% to 18.1%), showing that inclusion of these six studies did not inflate the pooled estimate.

Clinical variables not considered in this evaluation may be important determinants of the late-complication rate after short-course therapy for initially uncomplicated catheter-related S. aureus bacteremia. Data on elapsed time between onset of symptoms of infection and catheter removal was only given in one study [9]. Similarly, only one study provided information on the timing of initiation of antibiotic therapy in relation to onset of symptoms of infection [9]. In addition, only one study specified the method of screening for complications before or during therapy [5]. Two studies [7, 25] included patients with tunneled catheters, which may be associated with a different prognosis compared with those with short-term catheters. Differences in any of these variables might account for variation in reported late complication rates among the studies. Future investigators should include these factors in their study design.

Some believe that the major contribution of the studies we reviewed is finding that some patients with catheter-related S. aureus bacteremia do not develop late complications after short-course therapy. This information is only useful if this group of patients can be prospectively identified. To date, prospective identification has not been convincingly demonstrated. Several investigators have evaluated serologic methods for predicting which patients may safely be treated with short-course therapy. However, the ability of these tests to predict a low risk for late complications has varied considerably between studies [5, 14, 22, 35, 56].

We do not suggest that the idea of short-course therapy for uncomplicated catheter-related S. aureus bacteremia should be abandoned. On the contrary, this regimen should be studied further in a randomized trial as Iannini and colleagues originally suggested [3]. If the late complication rate proves to be acceptably low in all or a prospectively identifiable subset of patients with uncomplicated catheter-related S. aureus bacteremia, the shorter regimen should be implemented to reduce the costs and risks associated with prolonged therapy. If short-course therapy proves to have an unacceptable rate of late complications, an intermediate regimen suggested by Rahal [57], which includes 2 weeks of intravenous therapy followed by 2 weeks of oral antibiotics deserves study. Exactly what rate of late complication is considered to be acceptably low is likely to be a matter of debate. When more precise estimates are available, costbenefit analyses may help answer this question. However, even if such analyses show that it is cost beneficial to accept a certain excess percentage of complications after short-course therapy compared to conventional regimens there will likely be lack of complete agreement because the iatrogenic nature of this illness makes relapses and their complications particularly distressing. In any case, more accurate and precise information is needed before these questions can be properly addressed.

The accumulated data regarding the safety of short-course therapy are potentially flawed both by bias and by statistical imprecision. We therefore feel obliged to agree with Rahal [57] that until data from a well-designed, prospective study are available, patients should probably receive more than 2 weeks of therapy. The available data do not tell us what the optimal duration of therapy should be.