Standardized Treatment and Assessment Pathway Improves Mortality in Adults With Methicillin-resistant Staphylococcus aureus Bacteremia: STAPH Study

Abstract

Background

Methicillin-resistant Staphylococcus aureus (MRSA) bloodstream infection (BSI) management remains challenging for clinicians. Numerous in vitro studies report synergy when vancomycin (VAN) and daptomycin (DAP) are combined with beta-lactams (BLs), which has led to clinical implementation of these combinations. While shorter durations of bacteremia have often been reported, there has been no significant impact on mortality.

Methods

The Detroit Medical Center (DMC) developed and implemented a clinical pathway algorithm for MRSA BSI treatment in 2016 that included the early use of BL combination therapy with standard of care (VAN or DAP) and a mandatory Infectious Diseases consultation. This was a retrospective, quasi-experimental study at the DMC between 2013 and 2020. Multivariable logistic regression was used to assess the independent association between pathway implementation and 30-day mortality while adjusting for confounding variables.

Results

Overall, 813 adult patients treated for MRSA BSI were evaluated. Compared with prepathway (PRE) patients (n = 379), those treated postpathway (POST; n = 434) had a significant reduction in 30-day and 90-day mortality: 9.7% in POST vs 15.6% in PRE (P = .011) and 12.2% in POST vs 19.0% in PRE (P = .007), respectively.

The incidence of acute kidney injury (AKI) was higher in the PRE compared with the POST group: 9.6% vs 7.2% (P = .282), respectively. After adjusting for confounding variables including Infectious Diseases consult, POST was independently associated with a reduction in 30-day mortality (adjusted odds ratio [aOR], 0.608; 95% CI, 0.375–0.986).

Conclusions

Implementation of an MRSA BSI treatment pathway with early use of BL reduced mortality with no increased rate of AKI. Further prospective evaluation of this pathway approach is warranted.

Methicillin-resistant Staphylococcus aureus (MRSA) bloodstream infections (BSIs) are associated with significant mortality, morbidity, and increased health care expenditures [1, 2]. Several strategies have been previously investigated by our group to improve patient outcomes in MRSA BSI [3–5]. Combination therapy (CT) with a beta-lactam (BL) in MRSA BSI has been proposed as a treatment strategy due to multiple reports demonstrating in vitro synergy between vancomycin (VAN)/daptomycin (DAP) and BLs [6–10]. To date, studies evaluating combination VAN/DAP with BLs have shown shorter days of bacteremia, lower hospital stays, and reduction in infection recurrence but not evidence of mortality benefit [11–16]. In most of these evaluations, the BL was typically an “add-on” as an escalation of therapy in refractory infections or as part of an empiric therapy where the BL was used as empiric therapy for gram-negative infections and not purposely for the treatment of MRSA. The only prospective trials that have evaluated BL CT were the Combination Antibiotics for Methicillin Resistant Staphylococcus aureus (CAMERA)–1 and CAMERA-2 studies and the pilot study of DAP plus ceftaroline vs VAN [17–19]. Both studies found shorter days of bacteremia but no difference in mortality. The CAMERA-2 study was prematurely stopped due to high rates of acute kidney injury (AKI) in patients receiving VAN and flucloxacillin or cloxacillin [18]. While cefazolin was 1 of the BL options, there were too few patients receiving these antibiotics to draw any reliable conclusions [18, 20]. The DAP plus ceftaroline study, albeit limited with a small sample size, found a significant survival benefit among patients receiving the combination vs those receiving VAN alone, which led to early termination of the trial [19].

Based on the potential for improved outcomes with the use of BL CT, Infectious Diseases (ID) consultation, and microbiological assessment, a clinical pathway was developed at the Detroit Medical Center (DMC) with early use of BL CT as initial therapy for the treatment of MRSA bacteremia until at least 48 hours after blood culture sterilization (Figure 1) [21]. The primary BL was cefazolin; however, cefepime and other BL agents were allowed per patient specifics. Treatment modification may occur on days 3-5 of therapy and if necessary, followed by assessment on days 7–10. These were aimed to improve success rates with VAN or DAP, prevent the emergence of resistance, and reduce escalation to alternative, costlier, and more broad-spectrum agents [22]. The objective of this study was to evaluate the baseline characteristics and clinical outcomes of patients before and after the implementation of the MRSA BSI pathway at the DMC.

METHODS

STUDY Design and Population

We conducted a retrospective, quasi-experimental study at the DMC between December, 2013, and June 2020. The DMC is a single large health care system that includes 8 hospitals within the greater Detroit area of Michigan. The DMC operates 8 hospitals and institutes, including the Children’s Hospital of Michigan, Detroit Receiving Hospital, Harper University Hospital, Huron Valley-Sinai Hospital, Hutzel Women’s Hospital, Rehabilitation Institute of Michigan, Sinai-Grace Hospital, and DMC Cardiovascular Institute. Patients were screened and included if they (1) were age ≥18 years and (2) had ≥1 MRSA-positive blood culture meeting the Centers for Disease Control and Prevention Criteria for BSI [23]. Exclusion criteria are illustrated in Figure 2. Patients were classified in the prepathway (PRE) group if they were admitted before the pathway implementation date (ie, on or after September 1, 2016) or in the postpathway (POST) group if admitted after the pathway implementation date. Only patients’ first encounter was collected, and repeat encounters were excluded. This study was reviewed and approved by the WSU Institutional Review Board and the DMC Research Review Committee.

Patient Consent

Patient consent was not required for this retrospective analysis.

Data Collection and Study Definitions

Patients’ data were derived from the electronic medical record and entered into Research Electronic Data Capture (REDCap, Vanderbilt University). All blood cultures were processed at the DMC central microbiology laboratory according to standard procedures. MicroScan (Siemans Healthcare Diagnostics), Phoenix (BD) and Verigene (Luminex), and Etest (biomerieux) were utilized for antimicrobial susceptibility testing and/or bacterial identification depending on the time period. Variables associated with BSI were determined based on clinical notes and microbiological/diagnostic reports. The pathway is defined in detail in Figure 1, as well as in our institution’s portal [21]. Severity of illness and patient comorbidity were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) score and Charlson comorbidity index (CCI), respectively. Both were assessed at the time of index blood culture. Prolonged bacteremia was defined as lasting >120 hours [24]. Patients who died before clearing their bacteremia were excluded from the prolonged BSI analysis. In order to account for antibiotic adjustments during therapy, we describe several anti-MRSA/BL regimen scenarios in Table 2. Thirty-day and 90-day mortality were defined as death from any cause at 30 and 90 days, respectively. Sixty-day recurrence was defined as >1 MRSA-positive blood culture following clearance within 60 days of index blood culture collection. Safety outcomes are defined in Tables 1 and 3. Sources of bacteremia, occurrence of side effects, and other clinical variables were collected based on laboratory assessment and/or medical notes by the treating physician.

Outcome

The primary outcome was 30-day mortality. Secondary end points included 90-day mortality, 60-day recurrence, prolonged bacteremia, duration of bacteremia, hospital length of stay, ID consult, and AKI. All clinical outcome time points were measured from index blood culture collection.

Statstical Analysis

Descriptive statistics were used to evaluate patients’ demographics. Nominal data were reported as counts and percentages, and continous data were reported as medians and interquartile ranges (IQRs). Categorical variables were compared by the chi-square test, and continuous variables were compared by the Mann-Whitney U test. Multivariable logistic regression was used to assess the independent association between POST and 30-day mortality while adjusting for confounding variables. Data from POST, along with all variables associated with 30-day mortality in the bivariate analysis at a P value <.2, were entered into the model simultaneously and removed using a backward stepwise approach. Covariates were retained in the model if the P value for the likelihood ratio test for their removal was <.1. The variance inflation factor was used to assess the multicollinearity of covariates in the model, with values in the range of 1–5 considered acceptable. When certain variables were colinear, the variable with the highest number of patients was retained. The Hosmer-Lemeshow goodness-of-fit test was used to assess the model’s fit. All tests were 2-tailed, with P values <.05 considered statistically significant. If a patient died due to coronavirus disease 2019 (COVID-19) in POST, that patient was excluded from the regression analysis.

To account for inherent changes over the study time period such as in medical practice, strain epidemiology, and patient mix that may have influenced clinical outcomes, we performed an interrupted time series analysis. We examined changes over time as well as differential changes over time in the PRE and POST periods. Also, within the POST group, classification and regression tree (CART) analysis was performed to determine the time to start BL most predicitve of 30-day mortality. To assess the independent association between time to BL dichotomized at the CART-derived cut-point and 30-day mortality, a multivariable logistic regression model was performed. Additionally, we also repeated the multivariable logistic regression analysis using 90-day mortality as an outcome. IBM SPSS software, version 26.0 (SPSS, Inc., Chicago, IL, USA), was used for all calculations.

RESULTS

Overall, 1155 BSI patients were evaluated, and 342 were excluded. A total of 813 adult patients treated for MRSA BSI were evaluated (PRE, n = 379, and POST, n = 434) (Figure 2). The entire cohort was predominately male (75.4%), with a median (IQR) age of 50.5 (39.6–61.2) years. The median (IQR) APACHE II and CCI were 17 (11–23) and 3 (1–5), respectively. A comparison of baseline characteristics between PRE and POST is provided in (Table 1). Some notable differences were observed between the 2 groups. Diabetes was more prominent in the PRE compared with the POST group: 44.6% vs 36.2% (P = .015); as well as heart failure: 26.6% vs 19.6% (P = .017); and previous hospitalization: 40.9% vs 29.0% (P < .0001). Conversely, lack of comorbid conditions was less common in PRE compared with POST: 3.7% vs 9.2%, respectively (P = .002). Vancomycin susceptibilities over the 2 study periods are displayed in Table 1. The criteria of anti-MRSA agents and BL regimens in the PRE and POST groups are illustrated in Table 2. The most common first anti-MRSA agent is VAN in both PRE and POST: 90.0% and 95.9%, respectively (P = .001). Lack of BL CT (ie, monotherapy [MT]) was most common in PRE, 48.5%, while the most common first BL agent was cefepime in POST, 46.8%. The most common anti-MRSA agent at 48 hours was VAN in both PRE and POST: 70.2% and 85.7%, respectively (P < .001). Lack of BL combination was common in PRE, 63.3%, while the most common BL agent at 48 hours was cefazolin, 44.5%, in POST. The most common primary anti-MRSA agent was VAN in both PRE and POST: 53.0% and 71.2%, respectively (P < .001). Among patients who had a BL, the most common primary BL agent was cefepime in PRE, 16.5%, and cefazolin, 55.7%, in POST.

Thirty-day and 90-day mortality were higher in PRE compared with POST, 15.6% vs 9.7% (P = .011) and 19.0% vs 12.2% (P = .007), respectively (Figure 3). Sixty-day recurrence was comparable between PRE and POST: 5.8% and 4.3%, respectively (P = .978). In PRE, 24.5% of patients experienced prolonged bacteremia, compared with 21.8% in the POST group (P = .362). Regarding bacteremia duration, the mean (SD) was 4.2 (4.2) vs 3.6 (2.6) days in the PRE and POST groups, respectively (P < .001). The vast majority of patients had an ID consult, 92.7%, which was more common in POST compared with PRE: 94.5% and 90.8%, respectively (P = .042). The median (IQR) hospital length of stay was similar in the POST group compared with the PRE group: 11 (8–19) and 12 (8–19) days, respectively (P = .486). The incidence of AKI in the entire cohort was 8.3% and was higher in the PRE compared with the POST group: 9.6% and 7.2%, respectively, but was not statistically significant (P = .282). With CART analysis for the time to start a BL, none of the cutoff points identified by CART were predictive of the primary end point in the entire cohort.

A bivariate comparison of baseline criteria between patients with and without 30-day mortality is presented in Table 3. Notable differences between the 2 groups at the prespecified P value were included in the multivariable logistic regression model (Supplementary Table 1). Based on the final variables retained in the model, the pathway was independently associated with a reduction in 30-day mortality (adjusted odds ratio [aOR], 0.608; 95% CI, 0.375–0.986). Other variables independently associated with 30-day mortality are listed in Table 4. In the interrupted time series analysis when differential changes over time in the PRE and POST periods were examined for 30-day mortality as well as overall changes over time, the effect of time was not statistically significant (P = .710 and P = .404, respectively) (Figure 4). This indicates that 30-day mortality was not changing as a function of time and that the improvement in survival was due to POST intervention (Supplementary Figure 1). Additionally, when ID consult was excluded as a study variable and was considered as a component of POST in the logistic regression model, the primary analysis results remained consistent. Moreover, when the model was performed using 90-day mortality as the outcome with the same variable for the primary analysis, treatment within the pathway was also independently associated with reduced odds of 90-day mortality (aOR 0.634; 95% CI, 0.412–0.977). The Hosmer and Lemeshow test demonstrated an acceptable P value (P = .788).

DISCUSSION

As opposed to previous evaluations of BL CT, we initiated a formalized clinical pathway for the treatment of MRSA BSI. The pathway required early BL CT as a key component in addition to a mandatory ID consult, microbiological evaluation, and prespecified timely therapy assessments. Our study demonstrated a significant difference in both 30-day and 90-day mortality even after adjusting for confounding variables, including ID consultation. There are important differences worth highlighting from previous evaluations of BL combination. First the BL was initiated early in the treatment course as a function of the clinical pathway. Because early use of BL (ie, within 48 hours) was established in most patients (370, 88.9%), we were unable to determine a specific time to start BL that was most predictive of outcomes. This may have contributed to the improved mortality observed in our study. This was in contrast to CAMERA-2, where the average time to randomization was 48 hours [18]. This ultimately caused the majority of the MT arm to receive therapy within 72 hours of randomization, thus potentially misclassifying the MT arm toward the null. In addition, many of the CT patients did not receive any BL until 72 hours into their BSI culture, exceeding the window during which benefit is highly projected. Additionally, while ID consult is a known contributor to improved outcomes in MRSA BSI patients, we controlled for this factor to demonstrate that it was not the primary driver of the mortality benefit [25, 26].

Of interest, while we demonstrated that fewer patients experienced persistent bacteremia in the POST compared with the PRE group, this was not statistically significant. However, there was a statistically significant decrease in bacteremia duration, 4.2 vs 3.6 days in the PRE compared with the POST group, respectively. Although not all reports have not been able to demonstrate such an improvement, most studies demonstrated a statistically significant decrease in bacteremia by 1–2 days [11, 13–15, 17–19]. It is important to note that persistent bacteremia has been correlated with but not validated as a marker for mortality [27, 28].

Although there have been a few differences in underlying comorbid conditions between the PRE and POST groups, these factors did not play an impactful role on mortality as observed in our regression model. In addition, the clinical microbiology laboratory has used both the MicroScan (primarily PRE) and Phoenix (primarily POST) automated susceptibility platforms over the course of 9 years. MicroScan has been shown to be more likely to overcall an MIC value of 2 mg/L, whereas Phoenix tends to undercall an MIC of 2 mg/L [29, 30]. Therefore, it was not surprising to find more MIC values of 2 mg/L reported in PRE compared with POST, as Microscan was utilized in this time period. In addition, some isolates were also tested by Etest, which also by virtue of differences in inoculum tends to read higher than automated testing or the Clinical and Laboratory Standards Institute (CLSI) microdilution method. Of interest, our laboratory at Wayne State University frequently peforms vancomycin MICs for various research purposes using the “gold standard” broth microdilution technique on MRSA bloodstream ioslates from the Detroit Medical Center. We went back and were able to match 414 BSI isolates that were from patients in the PRE (110/379) and POST (314/434) time periods. In the PRE period, 90% of the VAN MICs were 1 or less, and in the POST period 95% were 1 mg/L or less. Therefore, there does not seem to be any major difference in vancomycin susceptibility between the 2 time periods. Overall, because of the inability to correctly identify VAN MIC of 1 mg/L and 2 mg/L in the entire cohort of patients, the exact impact of the VAN MIC on clinical outcome is unclear in this analysis.

AKI was relatively low in our cohort, even when BLs were combined with VAN. First, it is important to note that CAMERA-1 and CAMERA-2 used flucloxacillin, a semisynthetic penicillin associated with significant nephrotoxicity risks for the majority of the patients [17, 18]. In a secondary nephrotoxicity analysis of the CAMERA-2, AKI was independently more common in the flucloxacillin/VAN group but not the cefazolin/VAN group [31]. In our clinical pathway, cefepime (n = 282, 34.6%) and cefazolin (n = 149, 34.6%) were the most common BLs used in POST, suggesting that cephalosporin-based regimens appear to be safer when combined with VAN than penicillin-based regimens. A recent meta-analysis of CT studies also suggested that there is no difference in AKI between CT and MT, and notably most of the included studies had a cephalosporin-based BL [16]. Second, VAN AUC-guided dosing, which is associated with lower AKI, was implemented at the DMC in 2015 and is consequently more prevalent in POST compared with PRE [32, 33]. Third, our institution switched from piperacillin/tazobactam to cefepime as the primary empiric gram-negative agent of choice in 2015. As evident from the results, piperacillin/tazobactam was more prominent in PRE compared with POST (P < .001). Lastly, there was a lower proportion of patients on nephrotoxins in POST compared with PRE (P < .0001) [34]. Collectively, it is possible that the combination of these factors contributed to the lower incidence of AKI in POST, which has been previously demonstrated in studies conducted at our health care center [32, 35, 36].

Fewer patients were on DAP or ceftaroline as the primary anti-MRSA agent in POST compared with PRE: 24.7% vs 34.6% and 3.5% vs 11.9%, respectively. This demonstrates that we were able to improve success rates with VAN and decrease the use of costlier and more broad-spectrum agents.

Our study differs from previous evaluations of CT for the treatment of MRSA BSI as it is a real-world extensive evaluation of how a comprehensive pathway that incorporates diagnostics, timely ID consult, and systematic early utilization of CT can improve patient outcomes. While cefazolin is the mainstay of BL choice in the pathway, other BLs were also acceptable to pair with the anti-MRSA agent. We previously demonstrated that cefepime can positively impact patients’ outcomes in MRSA BSIs [13]. In order to account for a possible selection bias in patients with concomitant MRSA and gram-negative infection who had a higher anticipated mortality risk, we excluded those with polymicrobial bacteremia. Therefore, the impact of empiric cefepime as opposed to targeted BL agents was minimal. Additionally, because the study had a large sample size, particularly for a real-world analysis, we were able to detect and thus adjust for various confounding variables that may have contributed to positive clinical outcomes. To ensure that the results were not biased by PRE and POST variable imbalances, we repeated the logistic regression analysis with selected high-risk groups. In addition, we performed an inverse probability of treatment weighting (IPTW) analysis. The results of both of these analyses remained consistent with the primary findings and were statistically significant (Supplementary Table 2).

This study is not without limitations. First, this was a retrospective study and was challenged by inherent limitations of the design. Second, despite this being a multicenter study, it was limited to hospitals within a single health care system, and the results may not be generalizable to other patient populations. Additionally, because the time frame of the study was over 9 years, it may be possible that improvements in medical practices have contributed to patient outcomes as well as changes in strain epidemiology and virulence. However, we attempted to control for this by performing additional analyses using time in the study period as a variable, and the results remained consistent. Notably, we did find an increase in excess mortality during the last 4 months of our study period (ie, March to June of 2020), which may be related to the peak of the COVID-19 pandemic in the state of Michigan [37, 38]. Additionally, although we observed an improvement in 90-day mortality, this was likely impacted by improvement in 30-day mortality. Lastly, although our clinical pathway was reliant on CT with BLs, the impact of the pathway as an entire process might have been the driver of mortality rather than BL CT alone, and thus might explain the mortality benefit observed in our study but not in previous studies.

In conclusion, we have shown that a comprehensive clinical pathway to manage MRSA BSI can have a positive impact on patient outcomes, particularly improved survival. We demonstrated that the selection of the BL, such as cefazolin or cefepime, for CT is important as our results show that CT was safe and not associated with increased incidence of AKI. Lastly, while multiple anti-MRSA agents and BL were included in the clinical pathway, the predominant regimens were VAN/cefazolin and VAN/cefepime. Therefore, it would be of interest for future studies including prospective evaluations to be directed on evaluating anti-MRSA agents other than VAN in CT with these BLs.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Acknowledgments

Financial support. No funding or sponsorship was received for this study or for publication of this article.

Potential conflicts of interest. S.A., A.M.F., E.Z., T.M., S.H., J.J.Z., J.J., R.M., J.M.P., S.D., M.R.S., T.T., T.C., and N.R. have nothing to disclose; S.C.J.J. has received honoraria for speaking from Melinta and Sunovion; M.J.R.: research support, consultant or speaker for Allergan, Contrafect, Melinta, Merck, Motif, Paratek, Tetraphase, Shionogi, and Spero, and is partially supported by NIAID AI121400 and AI1300056-04.

Prior presentations. Data from a proportion of patients in this analysis were been presented, in part, at the American Society for Microbiology (ASM); October 3–7, 2018; San Francisco, CA (abstract 2379); at the Infectious Diseases Society of America (IDSA) Meeting; October 2–6, 2019; Washington, DC (abstract 2250); and in the following publications: Alosaimy et al., Jorgensen et al., and Zasowski et al. [11–13].

References

Author notes