Empirical antimicrobial therapy for bloodstream infections not compliant with guideline was associated with discordant therapy, which predicted poorer outcome even in a low resistance environment

Abstract

Objectives

To characterise all bloodstream infections (BSIs) in a low antimicrobial resistance (AMR) prevalence setting with regard to the appropriateness of empirical antimicrobial therapy, compliance with the national clinical practice guideline, de-escalation practice and outcome.

Methods

A retrospective observational study including patients aged $\ge$ 18 years admitted to a university hospital in central Norway with positive blood culture in 2019.

Results

We included 756 BSI episodes in our analysis. Empirical antimicrobial therapy was in accordance with the national guideline in 534 (70.6%), and not in accordance in 190 (25.1%) of the BSI episodes. There was a statistically significant association between compliance with the national guideline and concordant empirical antimicrobial therapy (p = .001). De-escalation of antimicrobial therapy was possible but not done in 217 (31.1%) of the BSI episodes. Variables identified as independent predictors of discordant empirical antimicrobial therapy included hospital department, type of empirical antimicrobial regimen, bacterial species, and AMR. Independent predictors of intra-hospital case fatality rate were coverage of empirical antimicrobial therapy, CCI-score, SAPS-II score, site of infection, and type of empirical antimicrobial regimen. Furthermore, the intra-hospital and long-term unadjusted all-cause case fatality rates were increased (p < .001, log-rank test for overall difference in survival) for the patients who received discordant empirical antimicrobial therapy.

Conclusion

Our study shows that empirical antimicrobial therapy initiated in accordance with national guideline recommendations increases the likelihood of receiving concordant therapy. Discordant empirical antimicrobial therapy was associated with poorer outcomes, even in a setting with low AMR prevalence.

Keywords:

• Bloodstream infection
• empirical antimicrobial therapy
• national guideline
• targeted antimicrobial therapy
• de-escalation
• antimicrobial stewardship

Introduction

Despite all advances in modern medicine bloodstream infection (BSI) is still a severe disease associated with high morbidity and mortality [1]. BSI is often associated with sepsis syndrome [2]. The outcome of BSI and sepsis is dependent on the prompt initiation of effective antimicrobial therapy [3,4]. Globally it is estimated that 31.5 million cases of sepsis are treated in hospitals each year, and sepsis contributes to an estimated 5.3 million deaths annually [5]. The lowest incidence of sepsis is found in Northern Europe, including Norway [5]. In Norway, it has been reported that sepsis contributed to 12.9% of hospital deaths in the period 2011–2012 [6].

Early administration of adequate antimicrobial therapy is a cornerstone in the initial management of sepsis [7]. Empirical antimicrobial therapy refers to the treatment administered before microbiological identification (ID) and antimicrobial susceptibility testing (AST), and such treatment is frequently based on clinical practice guideline recommendations, but also local empirical experience. The choice of empirical antimicrobial therapy should be broad enough to cover all likely pathogens, and therefore it might prove suboptimal for targeting the actual pathogen causing the infection [8]. Concordant or discordant therapy refers to whether or not the administered empirical antimicrobials cover the detected pathogen and the pathogen's antimicrobial susceptibility profile.

Previous studies have shown that early broad-spectrum antimicrobial therapy reduces mortality in patients with severe sepsis and septic shock, and this initial antimicrobial therapy is described as one of the most important treatments for sepsis [9]. However, the use of broad-spectrum antimicrobial therapy is also a known risk factor for the emergence of antimicrobial resistance (AMR) [10]. Early reporting of microbial ID and AST provides opportunities for early targeting and de-escalation of antimicrobial therapy which is associated with decreased antimicrobial consumption and costs [8]. De-escalation strategies have been shown to reduce the overall selection pressure for antimicrobial resistance, but have no overall effect on mortality [11,12].

The prevalence of AMR is low in Norway compared to most other countries. National surveillance on antimicrobial consumption and AMR in Norway in 2019, reported a prevalence of extended-spectrum beta-lactamase (ESBL) in Escherichia coli and Klebsiella pneumoniae isolated from blood culture of 7.1% and 5.7%, respectively [13]. The rate of resistance to gentamicin was 5.9% for E. coli and 4.4% for K. pneumoniae. The report also showed that 0.8% of Staphylococcus aureus isolated from BSI was methicillin resistant (MRSA), and 2.2% of Enterococcus species were resistant to vancomycin. The Norwegian national guideline recommends a combination of penicillin and gentamicin as empirical therapy for sepsis with unknown aetiology and infection focus [14]. This regimen covers the most common pathogenic bacteria causing BSI in Norway [13]. Recommended as alternatives to this regimen is piperacillin/tazobactam or cefotaxime.

The aim of this study was to investigate empirical antimicrobial therapy in patients with BSI admitted to a tertiary care university hospital in Norway in 2019, and proportions that were concordant or discordant. Furthermore, we wanted to describe whether the empirical antimicrobial therapy administered was compliant with the national guideline recommendations, to what extent the antimicrobial therapy was targeted and de-escalated after the AST report was received, and whether discordant empirical antimicrobial therapy affected the outcome.

Materials and methods

Study population and study design

The study population in this retrospective observational study consisted of patients aged $\ge$ 18 years admitted to St. Olavs Hospital, a tertiary care university hospital in Trondheim, Norway, with a positive blood culture collected in the period 01.01.2019 to 31.12.2019. All patients that had blood cultures drawn and that turned out positive during this period were evaluated for inclusion.

An episode was excluded from the study if the microbe isolated from the blood culture was not considered to be the cause of the clinical infection, there was fungal growth in the blood culture, the patient was declared terminally ill before antimicrobial therapy was started, if necessary information was missing, or the patient was transferred to another hospital before antimicrobial therapy was initiated. A study group, consisting of a medical microbiologist, an infectious disease specialist, and two medical students evaluated the AST report and concluded whether or not the isolated microbe was classified as a contaminant. Typically, blood cultures with coagulase-negative staphylococci, Corynebacterium spp, and Bacillus cereus were classified as contaminants and excluded from further analyses if there was no information in the electronic patient record indicating that the isolated microbe should be considered the cause of the current clinical infection.

Laboratory procedures

Blood cultures, normally two sets of one BACTEC Plus/F Aerobic and one BACTEC Plus/F Anaerobic bottle with up to 10 mL blood in each bottle, were collected and incubated in an automated blood culture system BD BACTEC FX (Becton Dickinson, BD Diagnostics, USA). Bacteria and yeast isolated were identified using Maldi-TOF MS on a Microflex LT mass spectrometer (Bruker Daltonics) with the MBT 7854 MSP Library, and other standard methods [15]. The standard EUCAST disc diffusion method was used to perform AST [16]. In addition, the agar gradient method with Liofilchem® MIC test strips (Liofilchem, Italy) was used for minimum inhibitory concentration (MIC) determination in selected cases. Interpretation of results was done according to breakpoints from EUCAST and NordicAST [17,18].

Definitions

A BSI episode was defined by a positive blood culture with a plausible pathogenic microorganism and concomitant clinical infection, as judged and documented by the attending physician. Positive blood culture was defined by the growth of one or more microbial species in one or more blood culture bottles. Further, a monomicrobial BSI episode was defined as growth in the blood culture of one microbial species during an episode of clinical infection, while a polymicrobial BSI episode was defined as the growth of a second microbial species in the same or subsequent blood culture bottles collected within 72 h of the first detected microbial species. A new BSI episode in the same patient was defined as growth of a new microbial species in a sample collected more than 72 h after the initial BSI episode, or if there was growth of the same microbial species in a new blood culture sampled more than 30 days after the initial BSI episode, with a period of no symptoms of infection in between [19,20].

The BSI episode was classified as hospital-acquired if the patient had been admitted to the hospital more than 48 h prior to blood culture collection [21]. A BSI episode was classified as procedure-related if the patient acquired the infection outside the hospital after having been subject to in-hospital procedures, and the clinician had concluded that it was due to this procedure. Typically, these procedures were prostate biopsies, transurethral resection of the prostate, and transurethral resection of bladder tumour. An episode was classified as community-acquired if it did not fall into either of the two first categories. Bacterial isolates belonging to the order Enterobacterales were classified as multidrug-resistant (MDR) if they were found to be resistant to three or more classes of antimicrobial drugs.

Comorbidity recorded in the electronic patient record at the time of admission was used to calculate the Charlson Comorbidity Index for each BSI episode [22–24]. The probability of survival in relation to each episode was scored according to the Severe acute physiology score (SAPS) II at the time when the blood culture was collected [25–27]. Since the information was lacking in the electronic patient journal for some of the variables, the calculation of the SAPS-II score was based on some modifications: PaO₂/FiO₂ data was not used for assessing the severity of the disease if the patient was on mechanical ventilation or CPAP (n = 39, 5.2%). In addition, variables with no data recorded in the electronic patient record were classified as normal based on the assumption that in the hospital routine clinicians commonly do not request data or analyses for functions where they do not suspect abnormal values and where they consider there is no clinical indication.

To describe the appropriateness of administered antimicrobial therapy, several methods were applied. The Norwegian national guideline for the appropriate use of antimicrobial therapy in hospitals was used to evaluate whether the initiated empirical regimen was compliant or not [28]. Empirical antimicrobial therapy was considered concordant or discordant depending on whether or not the patient received antimicrobial therapy that covered the detected pathogen based on ID and AST. If the microbe had not been tested for any of the given empirical antimicrobials, inferred values from antimicrobials within the same class were used, as recommended by EUCAST and NordicAST in 2019 [29]. If the AST report did not contain sufficient information for definite classification, the concordance of empirical antimicrobial therapy was classified as partial or uncertain. Targeted antimicrobial therapy was defined as any antimicrobial therapy given after the AST report was released by the microbiology department. Dosing of antimicrobial therapy was registered only for episodes with BSIs caused by S. aureus and treated with third-generation cephalosporins. In all other cases, dosing regimens were presumed to follow the standard regimen for each drug.

Data collection

Administrative, clinical, diagnostic, and therapy data were collected from the hospital's electronic patient records in a spreadsheet before analysis. The information on positive blood cultures was retrieved from the laboratory information system. The study group jointly concluded how the results should be classified in case of inconsistency in recorded data, or if the data should be classified as insufficient or inconclusive.

Statistical analyses

Binomial logistic regression with the calculation of odds ratios (ORs) with 95% confidence intervals (Cis) was used to investigate the associations between episode characteristics and discordance of empirical antimicrobial therapy, intra-hospital case fatality rate, and long-term case fatality rate. Fisher’s exact test was used to examine the association between compliance with the national guideline and concordance of empirical antimicrobial therapy. Log-rank test was used to compare the survival of the patients who received discordant empirical antimicrobial therapy versus the patients who received concordant empirical antimicrobial therapy. Statistical analyses were performed using IBM SPSS Statistics (Version 27.0.1.0 Armonk, NY: IBM Corp). A p-value < .05 was considered statistically significant.

Results

Episode characteristics

During the one-year study period, 924 patients had one or more blood cultures drawn that showed the growth of bacteria or fungi. Of these, 65 were excluded due to a lack of consent. Further, one or more exclusion criteria were met for 187 BSI episodes, as described in Figure 1. The remaining 756 BSI episodes were eligible for inclusion, of which 642 (84.9%) occurred as unique, single BSI incidents, and 114 (15.1%) as multiple BSI incidents with two or more BSI episodes.

Figure 1. Flowchart showing the inclusion process and selection of patients and episodes, and the number of episodes per patient.

Episode characteristics are presented in Table 1 and Table S1. The majority of patients with BSI episodes were admitted to a medical department (73.7%) while fewer were admitted to a surgical department (26.3%). The median age in this study was 74 years, with an interquartile range of 63–82 years. The most common comorbidity was diabetes mellitus. Penicillin allergy was reported in 5.2% of the patients.

Table 1. Characteristics of 756 episodes of bloodstream infection.

Sex	n (%)
Male	458 (60.6)
Age	n (%)
≤49	91 (12.0)
50–74	314 (41.5)
≥75	351 (46.4)
Comorbidities	n (%)
Diabetes mellitus
Uncomplicated diabetes mellitus	113 (14.9)
End-organ damage diabetes mellitus	18 (2.4)
Myocardial infarction	115 (15.2)
Transient ischaemic attack/cerebral vascular accident	103 (13.6)
Metastatic tumour	93 (12.3)
Congestive heart failure	86 (11.4)
Peripheral vascular disease	80 (10.6)
Connective tissue disease	72 (9.5)
Chronic kidney disease	68 (9.0)
Dementia	65 (8.6)
Localized tumour	65 (8.6)
COPD	51 (6.7)
Leukaemia	40 (5.3)
Peptic ulcer disease	37 (4.9)
Liver disease
Mild	14 (1.9)
Moderate to severe	9 (1.2)
Lymphoma	18 (2.4)
Hemiplegia	11 (1.5)
No underlying illness	203 (26.9)
Allergies	n (%)
Penicillin	39 (5.2)
Acquisition	n (%)
Community-acquired	592 (78.3)
Hospital-acquired	148 (19.6)
Procedure-related	16 (2.1)
Site of infection	n (%)
Urinary tract	252 (33.3)
Unknown	198 (26.2)
Abdomen	115 (15.2)
Respiratory tract	100 (13.2)
Skin/soft tissue	35 (4.6)
IV catheter	20 (2.6)
Endocarditis	14 (1.9)
Bone and joint	10 (1.3)
Other	12 (1.6)
CCI-score (median (interquartile range))	5 (3–7)
Modified SAPS-II score (median (interquartile range))*	26 (21–31)
Department	n (%)
Medical	557 (73.7)
Surgical	199 (26.3)
All-cause case fatality rate	n (%)
7-day case fatality rate	39 (5.2)
30-day case fatality rate	92 (12.2)
90-day case fatality rate	155 (20.5)

*SAPS-II score variables with missing values in the electronic patient record were classified as normal.

Microbe characteristics

Monomicrobial episodes constituted 693 of 756 (91.7%) BSIs. In 50 (6.6%) of the BSIs two different bacterial species were identified and in 13 (1.7%) three or more species were identified. The most frequently isolated bacterial pathogens were E. coli with 269 out of 756 (35.6%), followed by S. aureus with 129 out of 756 (17.1%) episodes, as shown in Table 2. Gram-negative, gram-positive and anaerobic pathogens constituted 443 (58.6%), 327 (43.3%) and 69 (9.1%) of the BSI episodes, respectively.

Table 2. Microbe characteristics.

Type of microbes	n (%)	AMR, n (%)	MDR*, n (%)
Gram negative bacteria
Escherichia coli	269 (35.6)	14 (5.2) ESBL	3 (1.1)
Klebsiella spp.
Klebsiella pneumoniae	62 (8.2)	6 (9.7) ESBL
Klebsiella oxytoca	17 (2.2)
Other	3 (0.4)
Pseudomonas spp.
Pseudomonas aeruginosa	25 (3.3)
Other	1 (0.1)
Enterobacter cloacae complex	17 (2.2)		1 (5.9)
Proteus mirabilis	12 (1.6)
Other	37 (4.9)
Gram positive bacteria
Staphylococcus spp.
Staphylococcus aureus	129 (17.1)	1 (0.8) MRSA
Other	7 (0.9)
Streptococcus spp.
Streptococcus pneumoniae	28 (3.7)
Streptococcus dysgalactiae	19 (2.5)
Streptococcus pyogenes	14 (1.9)
Streptococcus agalactiae	14 (1.9)
Streptococcus oralis	11 (1.5)
Enterococcus spp.		1 (2.0) VRE
Enterococcus faecalis	37 (4.9)
Enterococcus faecium	10 (1.3)
Other	2 (0.3)
Other	56 (7.4)
Anaerobic bacteria	69 (9.1)

*Three strains (two Escherichia coli and one Enterobacter cloacae complex) that were ESBL producers were also classified as MDR.

Among the E. coli and K. pneumoniae isolates, 5.2% and 9.7% were ESBL producers, respectively. Furthermore, of the 23 BSI cases with ESBL-producing bacteria, 3 strains (two E. coli and one Enterobacter cloacae complex) were also classified as MDR. S. aureus which was found to be methicillin resistant (MRSA) was isolated from one BSI episode (0.8%), and one of 49 (2.0%) enterococcal isolates was vancomycin-resistant (VRE).

Empirical antimicrobial therapy

Overall, 536 of 756 (70.6%) BSI episodes were prescribed empirical antimicrobial therapy in line with therapy recommendations in the national clinical practice guideline. On the other hand, 190 of 756 (25.1%) episodes were prescribed other regimens inconsistent with the guideline recommendations. For the remaining 32 BSI episodes, specific therapy recommendations were either lacking in the guideline (0.3%), the BSI episode was too complex to be ascribed to a specific therapy recommendation (0.3%), or the patient did not receive empirical antimicrobial therapy (3.7%).

The empirical antimicrobial therapy was discordant in 121 (16.0%) and concordant in 594 (78.6%) BSI episodes. For BSI episodes, in which empirical antimicrobial therapy was compliant with national guideline recommendations, concordant and discordant therapy was observed in 452 (85.8%) and 75 (14.2%) episodes, respectively (Figure 2). In contrast, for BSI episodes not compliant with the national guideline, 129 (73.7%) and 46 (26.3%) episodes received concordant and discordant therapy, respectively. There was a statistically significant association between compliance with the national guideline and concordant empirical antimicrobial therapy (Fisher’s exact test, p = .001).

Figure 2. Association between compliance with national clinical practice guideline (NCPG) and concordant empirical antimicrobial therapy. Of the episodes receiving empirical antimicrobial therapy compliant with NCPG, 452 (85.8%) and 75 (14.2%) were concordant and discordant, respectively. Of the episodes receiving empirical antimicrobial therapy not compliant with NCPG, 129 (73.7%) and 46 (26.3%) were concordant and discordant, respectively. Episodes with empirical antimicrobial therapy classified as either partial or uncertain coverage were excluded from the analysis. Asterisk indicates a statistically significant association between compliance with NCPG and concordant empirical antimicrobial therapy (Fisher’s exact test, p = .001).

Predictors of discordant empirical therapy were established using univariable and multivariable logistic regression analysis. Variables identified as independent predictors of discordant empirical antimicrobial therapy included hospital department, type of antimicrobial regimen, bacterial species, and antimicrobial resistance profile (Table 3 and Table S2). Third-generation cephalosporins in monotherapy were the strongest predictor of discordant empirical antimicrobial therapy among the empirical regimens, with a ten-fold increase in the rate of discordant therapy compared with narrow-spectrum penicillin combined with an aminoglycoside. Compared to narrow-spectrum penicillin combined with an aminoglycoside, all the other empirical antimicrobial regimens had an increase in the frequency of discordant empirical therapy.

Table 3. Predictors for discordant empirical antimicrobial therapy when compared with causative bacterial species and its susceptibility profile*.

		Univariable		Multivariable
Characteristic	No. of discordant (% within category)	Odds ratio (95% CI)	p-value	Odds ratio (95 % CI)		p-value
Acquisition
Community-acquired (n = 563)	83 (14.7)	Ref.		Ref.
Hospital-acquired (n = 136)	37 (27.2)	2.16 (1.39–3.37)	<.001	1.56 (0.88–2.77)		.131
Procedure-related (n = 16)	1 (6.3)	0.39 (0.05–2.96)	.359	0.43 (0.03–5.35)		.510
Overall p-value			.002			.244
CCI-score
0–3 (n = 217)	24 (11.1)	Ref.		Ref.
4–7 (n = 374)	68 (18.2)	1.79 (1.09–2.94)	.023	1.72 (0.95–3.11)		.072
8+ (n = 124)	29 (23.4)	2.46 (1.36–4.45)	.003	2.06 (1.00–4.24)		.049
Overall p-value			.010			.107
Department
Medical (n = 529)	103 (19.5)	Ref.		Ref.
Surgical (n = 186)	18 (9.7)	0.44 (0.26–0.75)	.003	0.42 (0.22–0.80)		.008
Empirical antimicrobial therapy
Narrow spectrum beta-lactam + aminoglycoside (n = 175)	9 (5.1)	Ref.		Ref.
Piperacillin-tazobactam (monotherapy) (n = 86)	12 (14.0)	2.99 (1.21–7.41)	.018	3.75 (1.27–11.12)		.017
3^rd generation cephalosporines (monotherapy) (n = 118)	25 (21.2)	4.96 (2.22–11.07)	<.001	9.91 (3.75–26.17)		<.001
Other antimicrobial therapy with only one regimen (n = 120)	31 (25.8)	6.42 (2.93–14.09)	<.001	6.99 (2.65–18.40)		<.001
Multiple empirical regimes (n = 216)	44 (20.4)	4.72 (2.23–9.97)	<.001	7.34 (3.04–17.73		<.001
Overall p-value			<.001			<.001
Compliant with national clinical practice guidelines for antimicrobial therapy
Not compliant with (n = 185)	46 (24.9)	Ref.		Ref.
Compliant with (n = 527)	75 (14.2)	0.50 (0.33–0.76)	.001	1.12 (0.65–1.92)		.682
Guidelines missing (n = 1)	0	0	1	0		1
Uncertain diagnosis (n = 2)	0	0	.999	0		.999
Overall p-value			.013			.983
Bacterial species
Escherichia coli (n = 232)	17 (7.3)	Ref.		Ref.
Klebsiella pneumoniae (n = 46)	2 (4.3)	0.58 (0.13–2.58)	.470	0.62 (0.12–3.21)		.570
Klebsiella oxytoca (n = 15)	1 (6.7)	0.90 (0.11–7.29)	.924	1.49 (0.17–13.14)		.720
Pseudomonas aeruginosa (n = 15)	4 (26.7)	4.60 (1.32–15.99)	.016	5.71 (1.44–22.67)		.013
Enterobacter cloacae complex (n = 14)	10 (71.4)	31.17 (8.97–111.50)	<.001	56.54 (12.80–249.78)		<.001
Proteus mirabilis (n = 9)	2 (22.2)	3.61 (0.70–18.76)	.126	7.14 (1.17–43.44)		.033
Staphylococcus aureus (n = 115)	37 (32.2)	6.00 (3.20–11.26)	<.001	9.69 (4.49–20.95)		<.001
Enterococcus faecalis (n = 23)	6 (26.1)	4.46 (1.56–12.80)	.005	11.58 (3.36–39.84)		<.001
Enterococcus faecium (n = 5)	2 (40.0)	8.43 (1.32–53.95)	.024	10.23 (1.37–76.56)		.024
Streptococcus spp. (n = 82)	0	0	.997	0		.997
Other monomicrobial (n = 103)	26 (25.2)	4.27 (2.20–8.30)	<.001	9.54 (4.20–21.65)		<.001
Polymicrobial (n = 56)	14 (25.0)	4.22 (1.93–9.20)	<.001	6.24 (4.46–15.77)		<.001
Overall p-value			<.001			<.001
Antimicrobial resistance
No resistance to empirical therapy (n = 693)	111 (16.0)	Ref.		Ref.
ESBL (n = 19)	10 (52.6)	5.83 (2.31–14.67)	<.001	50.92 (13.11–197.76)		<.001
VRE (n = 1)	0	0	1	0		1
MRSA (n = 1)	0	0	1	0		1
MDR (n = 1)	0	0	1	0		1
Overall p-value			.007			<.001
On antimicrobial therapy at the time when blood culture was collected
No (n = 602)	93 (15.4)	Ref.		Ref.
Yes (n = 113)	28 (24.8)	1.80 (1.12–2.92)	.016	1.36 (0.75–2.48)		.315

Univariable binominal logistic regression was used to find statistically significant dependent predictors. These covariates were used in a multivariable analysis to find independent predictors. The covariates that were not statistically significant in the univariable analysis are reported in Table S2.

*This table is a comparison between the BSI episodes who received discordant (n = 121) antimicrobial therapy with the BSI episodes who received concordant (n = 594) antimicrobial therapy. The BSI episodes who received partly concordant antimicrobial therapy (n = 7) or uncertain concordant antimicrobial therapy (n = 6) and the episodes who received no empirical antimicrobial therapy (n = 28) are excluded from this analysis.

Targeted antimicrobial therapy

The mean time from blood culture sampling to the finalisation of the AST was 2.65 days (95% CI 2.53–2.76, median 2 days). The antimicrobial therapy administered after the AST results were released (hereby termed targeted antimicrobial therapy) was concordant in 697 (92.2%) of the 756 BSI episodes. However, 13 of 756 (1.7%) BSI episodes still received discordant antimicrobial therapy. For the remaining BSI episodes, empirical antimicrobial therapy was either discontinued without targeted therapy (n = 15, 2.0%), antimicrobial therapy was initiated with uncertain efficacy due to lack of AST (n = 10, 1.3%), or the patient had deceased before targeted therapy (n = 21, 2.8%).

De-escalation from empirical antimicrobial therapy to a more narrow-spectrum targeted regimen according to the AST report, was done in 480 (68.9%) of 697 eligible BSI episodes. In contrast, in 217 (31.1%) of 697 eligible BSI episodes, broad-spectrum antimicrobials were continued although the AST result indicated that conversion to narrow-spectrum antimicrobial therapy would have been feasible.

Case fatality rate

Discordant empirical antimicrobial therapy was significantly associated both with increased intra-hospital (Figure 3A) and long-term (Figure 3B) unadjusted all-cause case fatality rate (p-value < .001, log-rank test for overall difference in survival). The intra-hospital case fatality rate was 19.0% versus 7.4% for patients receiving discordant or concordant empirical antimicrobial therapy, respectively. The median follow-up time for long-time survival was 577.5 days (interquartile range 156–718.5 days).

Figure 3. Kaplan-Meier plot showing the cumulative intra-hospital (A) and long-term (B) survival of patients with BSI who received concordant or discordant empirical antimicrobial therapy. BSI episodes with empirical antimicrobial therapy classified as uncertain coverage, partial coverage and no empirical antimicrobial therapy given were excluded from the analysis.

Coverage of empirical antimicrobial therapy, CCI-score, SAPS-II score, site of infection, and type of empirical antimicrobial regimen, were identified as independent predictors of intra-hospital case fatality rate (Table 4 and Table S3). The intra-hospital case fatality rate was about three times higher for patients receiving discordant empirical antimicrobial therapy, as compared to patients receiving concordant empirical antimicrobial therapy (OR 3.01, 95% CI 1.54–5.88).

Table 4. Predictors for intra-hospital case fatality rate.

		Univariable		Multivariable
Characteristic	No. of deaths (No. of death within category (%))	Odds ratio (95% CI)	p-value	Odds ratio (95 % CI)	p-value
Concordance of empirical antimicrobial therapy
Concordant (n = 594)	44 (7.4)	Ref.		Ref.
Discordant (n = 121)	23 (19.0)	2.93 (1.70–5.08)	<.001	3.01 (1.54–5.88)	.001
Partly (n = 7)	0	0	.999	0	.999
Uncertain (n = 6)	1 (16.7)	2.50 (0.29–21.87)	.408	0.57 (0.02–14.08)	.730
No therapy (n = 28)	3 (10.7)	1.50 (0.44–5.16)	.520	0.48 (0.09–2.64)	.400
Overall p-value			.005		.020
CCI-score
0–3 (n = 225)	5 (2.2)	Ref.		Ref.
4–7 (n = 396)	43 (10.9)	5.36 (2.09–13.74)	<.001	4.11 (1.45–11.62)	.008
8+ (n = 135)	23 (17.0)	9.04 (3.35–24.40)	<.001	4.585 (1.46–14.45)	.009
Overall p-value			<.001		.022
Modified SAPS-II score
<29 (n = 477)	23 (4.8)	Ref.		Ref.
29–39 (n = 216)	27 (12.5)	2.82 (1.58–5.04)	<.001	1.95 (1.01–3.78)	.047
40–51 (n = 55)	14 (25.5)	6.740 (3.23–14.09)	<.001	4.19 (1.70–10.35)	.002
≥52 (n = 8)	7 (87.5)	138.174 (16.31–1170.60)	<.001	112.40 (9.64–1303.41)	<.001
Overall p-value			<.001		<.001
Site of infection
Unknown (n = 198)	32 (16.2)	Ref.		Ref.
Urinary tract (n = 252)	9 (3.6)	0.19 (0.09–0.41)	<.001	0.223 (0.09–0.54)	<.001
Respiratory tract (n = 100)	16 (16.0)	0.988 (0.51–1.90)	.971	1.209 (0.54–2.70)	.642
Skin and soft tissue (n = 35)	5 (14.3)	0.865 (0.31–2.40)	.780	1.230 (0.36–4.25)	.744
Bone and joint (n = 10)	0	0	.999	0	.999
IV catheter (n = 20)	0	0	.998	0	.998
Abdomen (n = 115)	6 (5.2)	0.286 (0.12–0.71)	.007	0.419 (0.14–1.29)	.130
Other (n = 12)	2 (16.7)	1.038 (0.22–4.96)	.963	2.451 (0.32–18.86)	.389
Endocarditis (n = 14)	1 (7.1)	0.399 (0.05–3.16)	.384	0.696 (0.08–6.11)	.744
Overall p-value			.002		.023
Department
Medical (n = 557)	64 (11.5)	Ref.		Ref.
Surgical (n = 199)	7 (3.5)	0.28 (0.13–0.62)	.002	0.70 (0.28–1.73)	.439
Type of empirical antimicrobial therapy
Narrow spectrum beta-lactam + aminoglycoside (n = 178)	14 (7.9)	Ref.		Ref.
Piperacillin-tazobactam (monotherapy) (n = 88)	8 (9.1)	1.171 (0.47–2.91)	.733	0.513 (0.16–1.69)	.272
3^rd generation cephalosporines (monotherapy) (n = 119)	22 (18.5)	2.657 (1.30–5.43)	.007	0.869 (0.36–2.08)	.753
Other antimicrobial therapy with only one regimen (n = 123)	9 (7.3)	0.925 (0.39–2.21)	.860	0.235 (0.08–0.69)	.008
Multiple empirical regimens (n = 220)	15 (6.8)	0.857 (0.40–1.83)	.690	0.292 (0.12–0.72)	.007
No therapy (n = 28)	3 (10.7)	1.406 (0.38–5.24)	.612	0.161 (0.03–0.96)	.045
Overall p-value			.019		.014
Compliant with national clinical practice guidelines for antimicrobial therapy
Not compliant with (n = 190)	27 (14.2)	Ref.		Ref.
Compliant with (n = 534)	39 (7.3)	0.48 (0.28–0.80)	.005	0.523 (0.27–1.03)	.059
Guidelines missing (n = 2)	1 (50.0)	6.04 (0.37–99.43)	.208	11.907 (0.34–417.40)	.172
Unclear diagnosis (n = 2)	1 (50.0)	6.04 (0.37–99.43)	.208	0.395 (0.01–15.71)	.621
No therapy (n = 28)	3 (10.7)	0.72 (0.20–2.57)	.617	0.161 (0.03–0.96)	.045
Overall p-value			.012		.092

Univariable binominal logistic regression was used to find statistically significant dependent predictors, these covariates were used in a multivariable analysis to find independent predictors. The covariates that were not statistically significant in the univariable analysis are reported in Table S3.

Discussion

In this retrospective observational study of 756 BSI episodes over a one-year period at a tertiary care university teaching hospital in Norway with low AMR prevalence, we have shown that 70.6% received antimicrobial therapy compliant with national guideline recommendations. Of these, 14.2% were discordant. However, in BSI-episodes that received empirical antimicrobial therapy not compliant with the national guideline, 26.3% were discordant. We found a strong association between receiving a guideline-based regimen and concordant antimicrobial therapy. We have also reported that antimicrobial therapy was de-escalated, based on the AST report, to a more preferred regimen in 68.9% of BSI episodes while it was not de-escalated in 31.1% of BSI episodes. More importantly, we found that the intra-hospital and long-term case fatality rate was significantly higher for BSI episodes that received discordant therapy.

Surveillance of AMR prevalence is of paramount importance for the validity of therapy recommendations [30]. During the two decades where we have had national systematic countrywide AMR surveillance, the prevalence of important resistant pathogens like MRSA, ESBL-producing bacteria, and VRE have been low in Norway [13]. Based on decades of experience with clinical use combined with continued low AMR prevalence, narrow-spectrum beta-lactam therapy is still promoted in Norway for treatment of BSI and sepsis, often in combination with an aminoglycoside. The fact that 25.1% of included cases in our study were prescribed other regimens inconsistent with guideline recommendations, raises some concern. However, mechanisms that have impacted the prescribing habits among physicians were not a subject of this study.

We found a statistically significant association between compliance with guideline recommendations and concordant therapy. This finding supports that it is important to encourage the implementation of standardised therapy recommendations for specific clinical syndromes, as outlined in a previous antimicrobial stewardship guideline [31]. A recent systematic review of nine stewardship objectives from nearly 150 studies concluded that guideline-adherent empirical antimicrobial therapy was associated with a relative risk reduction for mortality of 35% [32].

Another finding is that 16.0% of BSI episodes received discordant empirical antimicrobial therapy. In a much larger American study, antimicrobial therapy was discordant in 19% of over 21.000 included BSI episodes [2]. In that study, infection with an AMR pathogen strongly predicted receiving discordant antimicrobial therapy. In our study, we also found that ESBL-producing Enterobacterales were independently associated with discordant therapy, although very few BSI episodes were due to ESBL-producing strains. Another predictor of discordant therapy was monotherapy regimens consisting of a third-generation cephalosporin or piperacillin-tazobactam, as compared to a narrow-spectrum beta-lactam in combination with an aminoglycoside. This is probably because we, according to EUCAST, classified the standard dosing of cefotaxime and ceftriaxone as insufficient to cover for S. aureus infections [17]. Of the 129 episodes with S. aureus infection in this study, 17 received a high dose of cefotaxime or ceftriaxone, while 6 received a standard dose. BSI episodes caused by Pseudomonas spp, Enterococcus spp, Enterobacter spp, Proteus spp, and as already mentioned S. aureus also predicted discordant therapy. This finding underscores that conditions other than AMR, including dosages of antimicrobial therapy, also need attention in order to reduce discordant therapy proportions.

In this study, we used clinical breakpoints for gentamicin as recommended by NordicAST in 2019. The gentamicin breakpoints were at that time classified as valid for all infections with Enterobacterales and staphylococci. In 2020 NordicAST changed its recommendation regarding gentamicin clinical breakpoints for Enterobacterales to be valid only for infections originating in the urinary tract. In addition, they stated that aminoglycoside treatment should be supported by other active treatments in case of systemic infections [33]. This study was not designed to compare results for gentamicin treatment between patients with infection origin in the urinary tract vs other site of infection. However, we believe that the results of this study are relevant despite these changes.

Early conversion from empirical to targeted antimicrobial therapy is uniformly recommended. This also applies to severe infections [34]. A recent meta-analysis concluded that antimicrobial monotherapy is non-inferior to duo therapy for severe infections, including sepsis and septic shock [35]. In our study, antimicrobial therapy was changed to pathogen-directed, targeted monotherapy with a more narrow-spectrum antimicrobial in 68.9% of eligible BSI episodes, whilst 31.1% were continued on broad-spectrum regimens. We were not able to identify reasons for deferring de-escalation strategies. Of note, 5.2% of included BSI episodes were labelled penicillin-allergic, but this can hardly explain the reasons for broad-spectrum non-targeted therapy.

Data in our study indicate that the case fatality rate was significantly increased among BSI episodes that received discordant therapy. This is consistent with several previous clinical studies [2,36–43] and reviews [34,44]. Unlike most similar studies, we applied inclusion criteria that did not select between specific pathogens or patient groups at risk but rather assessed all BSI episodes occurring over a 12-month period. The intra-hospital all-cause case fatality rate among BSI episodes receiving discordant or concordant empirical antimicrobial therapy in our study was 19.0% and 7.4% respectively. The fact that we observed such a difference in case fatality rate depending on concordance of empirical antimicrobial therapy even in a low AMR prevalence setting, is noteworthy and provides support for the importance of early concordant antimicrobial therapy.

Predictors of case fatality rate in multivariable analysis in our study were discordant empirical therapy, comorbidity, infection severity at presentation, site of infection, and choice of antimicrobial regimen. This is comparable to previous studies [2,36–43]. Of note, we were unable to assess clinical stages in sepsis, and BSI episodes presenting with especially pronounced clinical signs may have been allocated to broad-spectrum antimicrobial therapy. We therefore cannot rule out confounding by indication.

We consider it to be a strength of this study that the BSI episodes were identified based on laboratory-confirmed blood cultures systematically registered in the laboratory information system, and that the study included a high number of consecutive, unselected cases at a single centre tertiary care university teaching hospital in Norway where all types of inpatient departments and medical specialities are represented. However, there were also some limitations of this study. A subset of eligible patients still alive when the study started did not consent to participate and this may have led to the overrepresentation of diseased patients. Retrospective collection of clinical data depended on the availability and accuracy of such data in electronic patient records, and microbiology results from samples other than blood culture were not recorded in this study. Our practice of classifying missing values for the calculation of SAPS-II score as normal may have led to the underestimation of the SAPS-II score. Dosages of antimicrobial therapy were only available on scanned paper charts and therefore difficult to access. Such information was only registered for the use of third-generation cephalosporins in staphylococcal infections in this study. In addition, it is possible in a retrospective study that factors not recorded may have influenced the results obtained. Finally, as this is a single-centre study from a low antimicrobial resistance setting, whether the results from this study can be extrapolated to international settings needs further investigation.

In conclusion, this study showed that empirical antimicrobial therapy for BSI is more concordant with AST reports if such therapy is in accordance with national clinical practice guideline recommendations. The case fatality rate was significantly higher for BSI episodes that received discordant therapy. In this study, we disclosed predictors for discordant antimicrobial therapy. Antimicrobial stewardship programmes in general encourage health professionals and authorities to assess predictors for discordant empirical antimicrobial therapy carefully and implement antimicrobial stewardship measures to facilitate therapy prescribing. Measures should include strategies that target prompt identification and AST, guideline-based initiation of therapy, and compulsory early therapy assessments. Regardless of AMR prevalence, we encourage others to investigate concordant antimicrobial therapy in order to secure rigorous clinical guideline recommendations.

Ethical considerations

The Regional Committee for Medical and Health Research Ethics, health region mid-Norway (REK midt 210524) approved the study. Deceased patients at the start of study in June 2021 were included in the study without consent, while patients still alive received a written study invitation with opportunity to withdraw from the study. A data protection impact assessment was performed and approved by the hospital before start of study.

Author contributions

This study was conceived and designed by KG, EH, BW and JEA. KG, EH and BW acquired the data. KG, EH, BW and ØS analysed and interpreted the data. The study was supervised by BW and JEA, with input from JKD. KG and EH drafted the article, which was critically revised by BW, JKD, and JEA.

Acknowledgements

We express our sincerest gratitude towards Frode Width Gran for help with patient identification and Kjersti Haugum, Nina Bäcklund and the Department of Medical Microbiology, St. Olavs hospital for help with dispatch of invitation letters to the patients.

Disclosure statement

Author Bjørn Waagsbø is a representative and holds a 20% position to the directory group in the Norwegian Directorate of Health that seek to revise recommendations in the national clinical practice guideline for antimicrobial therapy.

Data availability statement

The datasets used and/or analysed during the current study are not publicly available due to restrictions from The Regional Committee for Medical and Health Research Ethics, health region mid-Norway (REK midt 210524).

Additional information

Funding

The study was financed by internal funds at Department of medical microbiology, St. Olavs hospital.