Usefulness of previous methicillin-resistant Staphylococcus aureus screening results in guiding empirical therapy for S aureus bacteremia.

BACKGROUND: Staphylococcus aureus bacteremia (SAB) is an important infection. Methicillin-resistant S aureus (MRSA) screening is performed on hospitalized patients for infection control purposes.
OBJECTIVE: To assess the usefulness of past MRSA screening for guiding empirical antibiotic therapy for SAB.
METHODS: A retrospective cohort study examined consecutive patients with confirmed SAB and previous MRSA screening swab from six academic and community hospitals between 2007 and 2010. Diagnostic test properties were calculated for MRSA screening swab for predicting methicillin resistance of SAB.
RESULTS: A total of 799 patients underwent MRSA screening swabs before SAB. Of the 799 patients, 95 (12%) had a positive and 704 (88%) had a negative previous MRSA screening swab. There were 150 (19%) patients with MRSA bacteremia. Overall, previous MRSA screening swabs had a positive likelihood ratio of 33 (95% CI 18 to 60) and a negative likelihood ratio of 0.45 (95% CI 0.37 to 0.54). Diagnostic accuracy differed depending on mode of acquisition (ie, community-acquired, nosocomial or health care-associated infection) (P<0.0001) and hospital (P=0.0002). At best, for health care-associated infection, prior MRSA screening swab had a positive likelihood ratio of 16 (95% CI 9 to 28) and a negative likelihood ratio of 0.27 (95% CI 0.17 to 0.41).
CONCLUSIONS: A negative prior MRSA screening swab cannot reliably rule out MRSA bacteremia and should not be used to guide empirical antibiotic therapy for SAB. A positive prior MRSA screening swab greatly increases likelihood of MRSA, necessitating MRSA coverage in empirical antibiotic therapy for SAB.

Staphylococcus aureus is a leading cause of bloodstream infections and is associated with a high mortality of 10% to 30% (1–4). Methicillin-resistant S aureus (MRSA) is highly prevalent. In Canada, it is estimated that 27% of all S aureus bacteremia (SAB) are MRSA; however, prevalence varies greatly depending on the region (5). MRSA bacteremia results in higher mortality, longer hospital stay and increased cost compared with methicillin-susceptible S aureus (MSSA) (6,7).

The decision to initiate an antibiotic with activity against MRSA in empirical therapy of suspected bacteremia is complex. For MRSA bacteremia, vancomycin is the standard antimicrobial agent, because β-lactam antibiotics are ineffective (8). Early empirical antibiotic therapy with MRSA coverage in MRSA bacteremia resulted in better clinical outcomes including lower mortality (9). However, vancomycin is inferior to β-lactam antibiotics in treating MRSA bacteremia (10,11). Furthermore, use of vancomycin increases antimicrobial resistance, such as vancomycin-resistant enterococci (12). Finally, use of vancomycin in the treatment of MRSA bacteremia is associated with risk for nephrotoxicity (8). In clinical practice, empirical vancomycin for suspected bacteremia is not universally used and not added unless there is an increased risk for MRSA (13–15). Ideally, empirical vancomycin is warranted only when the probability of MRSA is sufficiently high.

MRSA screening is usually performed in hospitals for infection control purposes. However, MRSA colonization may predict infection and, therefore, MRSA status may help guide empirical therapy in SAB (16,17). We conducted a retrospective cohort study to identify the clinical utility of past MRSA screening swab in predicting methicillin resistance for patients with SAB.

METHODS

Study design

The present study used data from a larger retrospective cohort study of SAB at six acute-care academic and community hospitals in the Greater Toronto Area (Ontario) from April 1, 2007 to April 1, 2010 (18). Research ethics board approval was obtained from each institution.

Consecutive patients were included in the study if they had ≥1 positive blood culture for S aureus and an MRSA screening swab was performed before blood culture susceptibility results. The MRSA screening swab may have been at a current or previous admission. Only the most recent MRSA screening swab was considered. Patients <18 years of age were excluded. Patients were only included in the study once, using the first positive blood culture as the index.

Data sources

Data were obtained from patients’ medical records at each site and entered into a standardized case report form. Variables collected from patient medical records included patient age, sex, hospital site, admitting service, date of MRSA screening swab collection, date and time of positive blood culture, date of admission and mode of acquisition.

MRSA screening procedure

MRSA screening criteria were similar for all sites according to Ontario provincial guidelines (19). Patients were screened if they satisfied any of the following criteria: any known history of colonization with antibiotic-resistant organism; any known history of contact with a patient known to be colonized with an antibiotic-resistant organism; admission to a health care facility in the past 12 to 24 months; or unable to provide information regarding any of the aforementioned risk factors.

For all sites, swab samples were obtained from the nares and rectum, as well as any insertion site or open wounds. Laboratory confirmation of MRSA was similar for all sites. MRSA swabs were incubated in MRSA selective media at 35°C or 37°C for approximately 24 h. Isolates were identified as S aureus if the rapid S aureus agglutination test and/or the tube coagulase test were positive. Methicillin resistance was confirmed using positive PBP2a agglutination testing and/or susceptibility testing (oxacillin screen plate or Vitek2 XL system AST-GP67 cards [Biomérieux, USA]) according to Clinical and Laboratory Standards Institute guidelines (20).

Blood culture procedure

Blood collection and culture were similar for all sites. Blood culture bottles were incubated at 35°C for a maximum of five days. Direct Gram staining was performed on all positive blood cultures, which were then subcultured onto blood agar plates. Blood agar plates were incubated at 35°C for two days. S aureus was identified when Gram staining showed Gram-positive cocci in clusters and tube coagulase test was positive. Methicillin resistance was determined by the same method used for MRSA screening.

Covariates

Potential covariates being considered included patient age (18 to 30, 31 to 45, 46 to 60, 61 to 75, 76 to 90 or >90 years of age), sex, hospital site, admitting service (medical, surgical or intensive care unit), time from MRSA screening swab collection to blood culture collection, time from admission to blood culture collection, MRSA screen with blood culture collection on same admission and mode of acquisition (community-acquired, nosocomial or health care-associated infection). The time from MRSA screening to blood culture collection was categorized into three groups: <2 days, two to 14 days and >14 days (21). Modes of acquisition, including community acquired, health care associated and nosocomial, were based on standard definitions (22).

Empirical antibiotic therapy

Empirical antibiotic therapy was defined as antibiotics started within three days of blood culture collection before full susceptibility results of the blood culture were known. Empirical MRSA coverage was considered to include intravenous vancomycin, quinupristin-dalfopristin or daptomycin.

Statistical analysis

Diagnostic properties of the MRSA screening swab were determined. MRSA screening was considered to be the test, and methicillin susceptibility of the first S aureus blood culture was considered the criterion standard. For sensitivity, specificity and predictive values, the 95% CIs were calculated using the Wilson method (23). In other analyses, when 2×2 diagnostic test tables had a cell with zero in it, 0.5 was added for calculation of diagnostic properties. For likelihood ratios, the 95% CI was also calculated (24).

Potential covariates were examined using a multivariable logistic regression model to identify variables related to differences in sensitivity and specificity, as described by Coughlin et al (25). In the logistic model, the dependent variable was the MRSA screen result and the methicillin susceptibility of the blood culture was entered as an independent variable along with other covariates. Potential covariates were as listed. Both sensitivity and specificity could be derived from the coefficients of the independent variables in the model (25). Several methods were used to confirm the significant covariates including univariate selection based on P value, full model with all covariates, as well as forward and backward stepwise regression based on the Akaike information criterion and likelihood ratio test.

To adjust for the significant covariates identified in the previous step (mode of acquisition and hospitals) from the multivariable logistic regression, the study population was stratified according to mode of acquisition. Within each subgroup, a random effects bivariate model (26) was used to calculate a summary estimate of sensitivity and specificity from the six different hospitals. The CIs for the likelihood ratios from the bivariate model were derived from a Monte Carlo simulation of 2000 samples.

Probability of post-test MRSA at different MRSA prevalence was calculated and plotted based on the pooled nonadjusted positive and negative likelihood ratios using the following formulas:

All reported CIs were two-sided 95% intervals and all tests were two-sided with a 5% significance level. All analyses were performed using R version 3.0.1 (R Foundation for Statistical Computing, Austria). Bivariate summary estimates of diagnostic properties were performed using R package mada.

RESULTS

There were 799 patients who underwent a MRSA screening swab before the susceptibility results of the initial positive blood culture were known (Table 1). Of all MRSA screening swabs, 448 (56%) were performed within two days of blood culture collection; 167 (21%) within two to 14 days; 182 (23%) within >14 days; and two had missing data. The minimum time from MRSA screening swab to the susceptibility results of the initial positive blood culture was one day. Given the minimum time of one day, and the fact that processing of MRSA screening swab took 24 h, all patients in the study were assumed to have MRSA screening swab results available before or at the same time as when the methicillin susceptibility results of the initial positive blood culture were known.

TABLE 1

Patient characteristics

Characteristic	All sites (n=799)
Age, years, median (interquartile range)	66.0 (52.0–79.0)
Male sex	501 (63)
Hospital site
A	121 (15)
B	102 (13)
C	223 (28)
D	167 (21)
E	68 (9)
F	118 (15)
Hospital admission service
Medical	498 (62)
Surgical	166 (21)
Intensive care unit	134 (17)
Other	1 (0.1)
Mode of acquisition
Community acquired	190 (24)
Health care associated	296 (37)
Nosocomial	297 (37)
Unable to determine	16 (2)
Susceptibility of initial Staphylococcus aureus positive blood culture
MRSA	150 (19)
Results of MRSA screening swab
Positive	95 (12)
Negative	704 (88)
Time course of MRSA screening to blood culture, days
MRSA screens performed before susceptibility report	799 (100)
MRSA screening to susceptibility report, median (IQR)	6.0 (3.0–18.0)
MRSA screening to culture collection, median (IQR)[*]	1.0 (0.0–12.0)
Culture collection to susceptibility report, median (IQR)	3.4 (2.9–5.2)

Data presented as n (%) unless otherwise indicated.

Data available for 797 patients. MRSA Methicillin-resistant S aureus

Diagnostic test characteristics

These results allowed for the determination of diagnostic test characteristics for the MRSA screen in predicting methicillin resistance of the initial positive blood culture (Table 2). Diagnostic test properties are shown for each hospital site in Appendix 1.

TABLE 2

Diagnostic properties of methicillin-resistant Staphlycoccus aureus (MRSA) screening in predicting methicillin susceptibility in S aureus blood culture

	Overall (n=799)	Community acquired (n=190)	Health care associated (n=296)	Nosocomial (n=297)
True positive[*], n	84	8	44	32
True negative†, n	638	169	232	223
False positive‡, n	11	1	8	1
False negative§, n	66	12	12	41
Sensitivity	56 (48–64)	40 (22–61)	79 (66–87)	44 (33–55)
Specificity	98 (97–99)	99 (97–100)	97 (94–98)	100 (98–100)
PPV	88 (80–93)	89 (57–99)	85 (73–92)	97 (85–100)
NPV	91 (88–93)	93 (89–96)	95 (92–97)	85 (80–88)
PLR (95% CI)	33 (18–60)	68 (9–516)	24 (12–47)	98 (14–706)
NLR (95% CI)	0.45 (0.37–0.54)	0.60 (0.42–0.86)	0.22 (0.13–0.37)	0.56 (0.46–0.69)
Bivariate sensitivity¶		43 (25–64)	74 (61–84)	49 (26–72)
Bivariate specificity¶		98 (94–99)	95 (92–97)	98 (94–99)
Bivariate PLR (95% CI)¶		18 (6–51)	16 (9–28)	21 (7–57)
Bivariate NLR (95% CI)¶		0.58 (0.38–0.77)	0.27 (0.17–0.41)	0.53 (0.29–0.77)

Data presented as % (95% CI) unless otherwise indicated.

Positive MRSA screening swab and MRSA blood culture;

Negative MRSA screening swab and MRSA blood culture

Positive MRSA screening swab and MRSA blood culture

Negative MRSA screening swab and MRSA blood culture

Bivariate summary estimate for each mode of acquisition. NLR Negative likelihood ratio; NPV Negative predictive value; PLR Positive likelihood ratio; PPV Positive predictive value

Hospital sites and mode of acquisition (community-acquired, nosocomial or health care-associated infection) were statistically significant covariates in the final multivariable logistic regression model (Table 3). Age, sex, admitting service, time from MRSA screen to blood culture collection, time from admission to blood culture collection, and MRSA screen with blood culture collection on same admission were not significant covariates. Univariate selection based on P value and stepwise regressions all derived the same model.

TABLE 3

Final logistic regression model* for probability of a positive methicillin-resistant Staphylococcus aureus (MRSA) screening test

Covariate	OR (95% CI)	P	P[†]
MRSA status on blood culture	197.47 (84.71–527.55)	<0.0001	<0.0001
Hospital site			0.0002
D	Reference
A	0.99 (0.34–2.76)	0.9796
B	0.57 (0.17–1.82)	0.3510
C	0.25 (0.09–0.65)	0.0049
E	0.06 (0.01–0.29)	0.0008
F	0.20 (0.05–0.72)	0.0172
Mode of acquisition			<0.0001
Community	Reference
Health care associated	6.94 (2.38–22.55)	0.0007
Nosocomial	1.13 (0.39–3.43)	0.8228

The logistic model used MRSA swab result as the dependent variable. The independent variables included methicillin susceptibility of blood culture along with the covariates listed above. CIs are likelihood ratio-based CIs. Both sensitivity and specificity can be derived from the coefficients of the model, as described by Coughlin et (26). The model used listwise deletion and included 780 patients with no missing data;

Values of the likelihood ratio test

For each mode of acquisition, a bivariate summary estimate of sensitivity, specificity and likelihood ratios were calculated from the six sites (Table 2).

The utility of MRSA screening results was modelled for different prevalences of MRSA by plotting post-test probability of MRSA, based on observed pooled and unadjusted positive and negative likelihood ratios (Figure 1).

Figure 1)

Post-test probability of methicillin-resistant Staphylococcus aureus (MRSA) at different MRSA prevalence in S aureus bacteremia. NLR Negative likelihood ratio; PLR Positive likelihood ratio; SAB S aureus bacteremia

Empirical MRSA coverage

Of 150 patients with MRSA bacteremia, 59 (39%) had empirical MRSA coverage and 91 (61%) did not. Of 91 patients with MRSA bacteremia with inappropriate empirical antibiotic therapy, 42 (46%) had a positive prior MRSA screening swab.

DISCUSSION

Our multicentre retrospective cohort study at six acute-care academic and community hospitals examined consecutive patients with SAB. From the 799 patients studied, we found that the MRSA screening swabs preceded the susceptibility results of the initial positive blood culture by a median of six days, which may have helped guide empirical antibiotic therapy.

Mode of acquisition was an important covariate for the diagnostic accuracy of the MRSA screen. The positive likelihood ratio for mode of acquisition was high, ranging from 16 to 21, regardless of how the infection was acquired. A positive likelihood ratio >10 is considered to be clinically helpful (27). Therefore, a positive MRSA screening swab may help guide treatment because the risk of methicillin resistance for SAB is increased markedly. The negative likelihood ratio ranged from 0.27 to 0.58 for different modes of acquisition. A negative likelihood ratio from 0.2 to 0.5 makes a small change to the probability of disease (27). Therefore, a negative MRSA screening swab result is not useful in ruling out MRSA bacteremia.

There are few studies investigating the diagnostic accuracy of MRSA screening. The majority examined diagnostic accuracy of MRSA screening in the context of all clinical infections including non-S aureus infections (28,29). In contrast, we examined the diagnostic accuracy of MRSA swabs in SAB. One study by MacFadden et al (21) examined diagnostic accuracy of MRSA swabs in all S aureus infections including nonbacteremic infections. The overall specificity in our study was similar to their study, although our overall sensitivity was lower. The differences may be attributed partially to the types of clinical isolates included. Their study included isolates from both sterile and nonsterile sites that were treated with antistaphylococcal antibiotics, which could include colonization samples that were not clinical infections. In comparison, our study only included S aureus-positive blood cultures as clinical infections. As well, MacFadden et al (21) reported results from a single academic centre whereas the present study involved a diverse group of six academic and community hospitals. One of our study sites was also where MacFadden et al conducted their study, but there was no overlap of data between the two studies.

The present study had several strengths, including its size as the largest study, in examining the diagnostic accuracy of MRSA screening swabs in SAB. The study was conducted across many sites, both academic and community hospitals, enhancing its generalizability. In addition, methicillin susceptibility of S aureus determined from blood culture was an appropriate and independent standard that was uniformly performed in all patients regardless of MRSA screening swab results. Finally, the inclusion of only blood cultures growing S aureus, reflecting sterile site growth, ensured that all infections were true clinical infections and not colonization.

The present study had several limitations that merit discussion. First, as a retrospective chart review in which assessors of MRSA screening swab results were not blinded to the MRSA blood culture results, there could be potential information bias. However, the laboratory tests were determined and reported independent of other test results, making this unlikely. Second, there appeared to be heterogeneity in terms of diagnostic properties among MRSA hospital sites. The heterogeneity among different hospital sites might have been a result of minor differences in MRSA screening criteria, patient population at hospital sites and different rates of intrahospital MRSA transmission. However, the bivariate summary estimate accounts for this difference among hospital sites and provides a more conservative estimate of diagnostic properties. Moreover, the heterogeneity exists mainly in sites with fewer patients where corrections of adding 0.5 to all cells needed to be made. Sites with a greater number of patients had more consistent results. Finally, previous data have shown that different MRSA genotypes have a different probabilitites of causing SAB. Unfortunately, we did not collect any information on MRSA genotypes. However, MRSA genotype may not be clinically important, in that treatment of MRSA is the same regardless of MRSA genotypes in clinical practice and MRSA genotype is not considered in hospital infection control practices currently.

Our study identified the mode of acquisition as a possible significant covariate for diagnostic accuracy of MRSA screening. It may be that health care-associated infections were mostly associated with intravenous and hemodialysis therapy, and that these infections were strongly associated with pre-existent MRSA colonization. In contrast, for community-acquired and nosocomial infections, patients were more likely to be newly colonized with MRSA and, thus, were less likely to be detected by prior MRSA screening. Unlike the previous study (21), time from MRSA screening to blood culture collection was not a significant predictor. This is likely due to the fact that events such as mode of acquisition of infection and hospitalization play a more significant role in MRSA colonization and infection than the length of time. Mode of acquisition was not considered in the previous study (21).

Our study results have important clinical implications. A negative MRSA screen result does not significantly decrease the probability of MRSA bacteremia. The poor sensitivity and low negative likelihood ratio results in many false negatives. In these false-negative cases with MRSA bacteremia, withholding antibiotics with activity against MRSA based on negative MRSA screening swab may greatly increase the risk for mortality (9). Therefore, a negative MRSA screening swab is not useful and should not be considered in the decision of MRSA coverage for empirical antibiotic therapy. On the other hand, a positive prior MRSA screen rules in MRSA and empirical intravenous antibiotics with activity against MRSA should be started. In our cohort, 91 of 150 (61%) patients with MRSA bacteremia did not receive empirical MRSA coverage based on clinical judgment. In these 91 patients, the rule of a positive MRSA screen necessitating anti-MRSA empirical coverage would result in empirical MRSA coverage being added to 42 (46%) of the 91 MRSA bacteremia cases in which empirical MRSA coverage was originally missed. Following the same rule, MRSA empirical coverage would only be added to 11 (2%) of the 649 patients with MSSA bacteremia. Thus, this rule would significantly increase the rate of correct empirical MRSA coverage in MRSA bacteremia cases while adding minimal unnecessary anti-MRSA empirical coverage in MSSA bacteremia cases.

Our findings demonstrate that screening tests for infection control purposes can provide valuable, clinically relevant information for making treatment decisions.

Characteristic	Hospital site
	A (n=121)	B (n=102)	C (n=223)	D (n=167)	E (n=68)	F (n=118)
Age, years, median (Interquartile range)	65 (46–80)	76 (63–84)	67 (55–78)	59 (49–70)	70 (56–79)	63 (53–78)
Male sex	71 (59)	52 (51)	144 (65)	119 (71)	44 (65)	71 (60)
Hospital admission service
Medical	80 (66)	74 (73)	116 (52)	110 (66)	44 (65)	74 (63)
Surgical	22 (18)	11 (11)	52 (23)	37 (22)	12 (18)	32 (27)
Intensive care unit	19 (16)	17 (17)	54 (24)	20 (12)	12 (18)	12 (10)
Other	0 (0)	0 (0)	1 (0.5)	0 (0)	0 (0)	0 (0)
Mode of acquisition
Community	32 (26)	30 (29)	42 (19)	30 (18)	20 (29)	36 (31)
Health care associated	43 (36)	42 (41)	80 (36)	80 (48)	23 (34)	28 (24)
Nosocomial	45 (37)	29 (28)	93 (42)	55 (33)	22 (32)	53 (45)
Unable to determine	1 (1)	1 (1)	8 (4)	2 (1)	3 (4)	1 (1)
MRSA on blood culture	28 (23)	19 (19)	51 (23)	24 (14)	13 (19)	15 (13)
Positive MRSA screening	22 (18)	16 (16)	25 (11)	23 (14)	3 (4)	6 (5)
Diagnostic test
True positive, n	20	13	21	21	3	6
True negative, n	91	80	168	141	55	103
False positive, n	2	3	4	2	0	0
False negative, n	8	6	30	3	10	9
Sensitivity, % (95% CI)	71 (53–85)	68 (46–85)	41 (29–55)	88 (69–96)	25 (10–51)	41 (21–64)
Specificity, % (95% CI)	98 (93–99)	96 (90–99)	98 (94–99)	99 (95–100)	99 (92–100)	100 (96–100)
PLR (95% CI)	33 (8–134)	19 (6–60)	18 (6–49)	63 (16–250)	28 (2–511)	85 (5–1429)
NLR (95% CI)	0.29 (0.16–0.53)	0.33 (0.17–0.64)	0.60 (0.48–0.76)	0.13 (0.04–0.37)	0.76 (0.56–1.03)	0.60 (0.40–0.90)

Data presented as n (%) unless otherwise indicated. MRSA Methicillin-resistant Staphylococcus aureus; NLR Negative likeihood ratio; PLR Positive likelihood ratio