Aminoglycoside versus β-lactam treatment for urosepsis—a retrospective cohort study

Abstract

Objectives

Aminoglycosides, once less favoured compared with β-lactams (BLs) for treating urinary tract infections (UTIs), have gained attention due to the rising prevalence of ESBL-producing Enterobacterales. However, comparative data on the efficacy and safety of aminoglycosides versus BLs are limited.

Methods

We retrospectively compared patients with Gram-negative bacteraemic UTIs, who received monotherapy with aminoglycosides or BLs for at least 3 days. The primary outcome was clinical improvement at 72 h. Secondary outcomes included clinical improvement by discharge, time to improvement, mortality, length of stay, relapse rate and kidney injury.

Results

Out of 134 patients, 96 received BLs and 38 received aminoglycosides. BL recipients had more comorbidities, renal failure and higher clinical severity of bacteraemia. Clinical improvement for BLs versus aminoglycosides was similar at 72 h (55% versus 65.8%, P = 0.335) and by discharge (87.5% versus 94.7%, P = 0.663). A multivariate analysis accounting for baseline differences showed similar efficacy at both timepoints, with ORs for improvement for BLs versus aminoglycosides of 1.52 (95% CI 0.54–4.31). Hospital stay was 1.7 days shorter with aminoglycosides. Other secondary outcomes were not different between groups. Kidney injury was more common with aminoglycosides, but this difference was not significant and was not found when the analysis was limited to patients with a creatinine level of <1.5 mg/dL.

Conclusions

In this cohort of bacteraemic patients with UTI, aminoglycosides were associated with similar outcomes as BLs, with no significant risk of toxicity. Given their broad Gram-negative coverage and favourable pharmacokinetics, aminoglycosides could be reconsidered as a first-line treatment option for UTIs.

Introduction

Urinary tract infections (UTIs) are among the most common indications for inpatient antibiotic treatment. Complicated UTIs (cUTIs), including febrile UTI and urosepsis, are usually treated with IV antibiotics until clinical improvement. Empirical treatment options should cover Gram-negative bacteria, most commonly Enterobacterales, and include β-lactam (BL) antibiotics (penicillins, cephalosporins and carbapenems), quinolones and aminoglycosides (AGs).^1,2 The use of sulphonamides is hindered in many areas by high resistance rates.

Once a first-line parenteral option for UTI, AGs have mostly been replaced by BLs (mainly third-generation cephalosporins), owing to the superior safety profile of the latter class. Nevertheless, empirical use of BLs is increasingly challenged by the emergence of ESBL-producing Enterobacterales (ESBL-PE). Hence, broader-spectrum BLs such as piperacillin/tazobactam, carbapenems and novel antibiotics such as ceftazidime/avibactam, meropenem/vaborbactam and others, are being used more frequently, with the risk of driving resistance even further.

AGs offer significant advantages for the treatment of UTI, including: a broad spectrum against Gram-negative organisms; relatively low rates of resistance; little perturbation of the gut microbiota; favourable pharmacokinetics, including high and prolonged concentrations in the renal parenchyma and urine; and once-daily dosing owing to a post-antibiotic effect. Despite that, AGs are not commonly used in many hospitals due to fear of nephrotoxicity and ototoxicity, and due to the lack of quality evidence on non-inferiority compared with BLs, especially in bacteraemic patients. Of note, current practice guidelines list AGs as a treatment option for pyelonephritis and cUTI, although with a lower level of evidence and recommendation.^1–3

The Samson Assuta Ashdod University Hospital is a 300-bed teaching hospital. The resistance rate of Enterobacterales is high, with 30%–50% of uropathogens being ESBL-PE (unpublished data, A. Cahan). Our institutional antimicrobial guidelines recommend AGs (gentamicin or amikacin) as first-line treatment of cUTI for patients with an estimated glomerular filtration rate (GFR) greater than 30 mL/min. Alternatives in patients with low GFR include ceftriaxone, or piperacillin/tazobactam in patients with a high likelihood of having an infection caused by ESBL-PE. Following these guidelines is at the discretion of treating physicians, who may opt for another treatment as they see fit.

We retrospectively studied the outcomes of patients with UTI treated with either AGs or BLs. In order to select patients with a definite diagnosis of UTI, and avoid cases of asymptomatic bacteriuria, we limited the study to patients with bacteraemia. We hypothesized that treatment with AGs or BLs would be associated with similar outcomes.

Methods

Study design

The study was an observational retrospective comparison of hospitalized patients with Gram-negative bacteraemic UTI, empirically treated with either AGs or BLs for at least 3 days. Electronic medical records of patients admitted between November 2017 and March 2022 were retrospectively reviewed. Consecutive patients 18 years old or older who had at least one positive blood culture growing Gram-negative bacilli, and a urine culture growing the same species between 1 day before and 1 day after the culprit blood culture, were screened for inclusion. Antibiotic treatment during the first 72 h was recorded, and patients were grouped according to the first appropriate antibiotic given based on in vitro susceptibility testing. We included only patients who received either an AG or a BL as monotherapy for the first 72 h of admission. Excluded were: patients with polymicrobial bacteraemia; those with an alternative diagnosis; patients discharged earlier than 72 h after admission; and patients treated with antibiotics other than a BL or AG. Patients could be given more than one type of AG or BL sequentially, but not a combination of both (e.g. meropenem empirical treatment with down-escalation to ceftriaxone after susceptibility testing results). Demographic, clinical and laboratory data were collected from patients’ electronic files.

Outcomes

The primary outcome was clinical improvement by 72 h following admission. Patients were considered to have clinical improvement if they were alive, had a body temperature of <37.8°C during the preceding 24 h, were haemodynamically stable, had documentation in their chart of significant improvement of local symptoms; had a WBC count of <12 × 10³ cells/μL; and were not oliguric (except for chronic oliguria in patients with end stage renal disease). Secondary outcomes included: (i) clinical improvement by the end of admission using the same definitions as for the primary outcome, except for the WBC count, which was not used (since in many cases, a blood count was not repeated just before discharge); (ii) time to improvement, defined as the number of hours until the above-mentioned criteria for improvement were met; (iii) in-hospital mortality attributable to the infection; (iv) length of stay; (v) relapse within 90 days, defined as bacteruria recurring with the same strain; (vi) treatment-associated kidney injury assessed using the ratio of serum creatinine on Day 3 to Day 1 of treatment, and the percentage of patients whose creatinine had increased by 50% or more on Day 3.

Comorbidities were assessed according to the Charlson comorbidity score (CCI).⁴ The severity of the infection was assessed with the Pitt bacteraemia score.⁵

Statistical analysis

Baseline characteristics of patients in the BL and AG groups were compared using Student’s t-test or the Kruskal–Wallis test, as applicable. ORs for laboratory and clinical outcomes by empirical antibiotic treatment group were computed. Multivariate analysis was done using logistic regression with either clinical improvement after 72 h or by the end of admission as the outcome. Parameters were entered into the regression model if they were significantly different between groups or were clinically thought to be predictive of improvement (including the presence of urinary obstruction, urological procedures for source control and the time to appropriate treatment).

A selection model for propensity score weighting was used to account for differences in the distribution of confounders between treatment groups. Propensity scores were assigned using logistic regression with age, CCI, Pitt bacteraemia score, serum creatinine, and C-reactive protein (CRP) as covariates. Propensity scores were then used as weights in a weighted logistic regression.

As the institutional protocol recommends against AG use in patients with significant renal failure, a sensitivity analysis was performed, including patients with a creatinine level of ≤1.5 mg/dL at presentation. Statistical analysis was done using R version 4.3.2.^6–9

Ethics

The study was approved by the institutional review board (#0043-19-AAA). Due to the retrospective and observational nature of the study, informed consent was waived.

Results

Between November 2017 and March 2022, there were 332 patients with Gram-negative bacteraemia and the same organism isolated from a urine culture. Figure 1 shows a flow chart of patient selection for this study. Seventy-two patients were excluded because of polymicrobial bacteraemia, having a non-urinary source of infection, discharge before 72 h or not receiving any appropriate treatment. Of the remaining 260 patients, 121 received combination therapy or agents other than BLs or AGs, and 5 had missing laboratory data, leaving a study population of 134 patients, 96 of whom were treated with a BL and 38 with an AG (25 and 13 with gentamicin and amikacin, respectively). Baseline characteristics and outcomes of the patients included are shown in Table 1.

Figure 1.

Study population flow chart.

Table 1.

Characteristics of bacteraemic patients with cUTIs treated with BLs or AGs

Parameter	BL group	AG group	P value
N	96	38
Male sex, n (%)	39 (40.6)	16 (42.1)	0.876
Age (years), median (IQR)	79.2 (69–85.5)	71.6 (62.2–82.0)	0.073
Long-term care facility residency, n (%)	18 (18.7)	5 (13.2)	0.440
CCI, median (IQR)	6 (4–7)	4 (2–6)	<0.001
Admission in the previous 6 months, n (%)	26 (27.1)	10 (26.3)	0.928
Permanent urinary catheter or nephrostomy, n (%)	10 (10.4)	5 (13.2)	0.651
Diabetes mellitus, n (%)	45 (46.9)	16 (42.1)	0.583
Advanced renal failurea, n (%)	40 (41.7)	0 (0)	<0.001
Pitt bacteraemia score, median (IQR)	2 (0–3)	0.5 (0–1)	0.023
Any urinary obstructionb, n (%)	30 (31.2)	7 (18.4)	0.136
Baseline creatinine at presentation (mg/dL), median (IQR)	1.8 (1.17–2.6)	1.0 (0.7–1.2)	<0.001
Baseline CRP (mg/L), median (IQR)	183 (94–270)	140 (61–179)	>0.001
Any urological source controlc, n (%)	46 (47.9)	14 (36.8)	0.248
Time to appropriate antibiotic treatment (h), median (IQR)	2.9 (1.03–8.47)	3.48 (1.51–15.48)	0.295
Outcomes
Clinical improvement at 72 hd, n (%)	55 (57.3)	25 (65.8)	0.335
Clinical improvement at the end of hospital staye, n (%)	84 (87.5)	36 (94.7)	0.663
Time to improvement (h), median (IQR)	72 (48–96)	61.9 (35.7–96)	0.376
In-hospital death, n (%)	12 (12.5)	2 (5.3)	0.221
Hospital stay (days), median (IQR)	5.9 (4.2–8.8)	4.6 (3.5–7.5)	0.01
Relapse of UTI within 90 days, n (%)	14 (14.6)	3 (7.9)	0.298
Time to relapse (days), median	60.3	61.7	0.899
>50% increase in creatinine at Day 3, n/N (%)	2/91 (2.2)	3/35 (8.6)	0.130
Creatinine Day 3/Day 1 ratio	0.778	0.895	0.012

^aBaseline creatinine ≥ 3 mg/dL.

^bIncluding urinary retention and hydronephrosis.

^cIncluding insertion of urinary catheters, nephrostomy, ureter catheter etc.

^dDefined as the patient was alive, had a body temperature of <37.8°C during the preceding 24 h, was haemodynamically stable, had documentation in their chart of significant improvement of local symptoms, had a WBC count of <12 × 10³ cells/μL, and was not oliguric (except for chronic oliguria in patients with end-stage renal disease).

^eDefined as the patient was alive, had a body temperature of <37.8°C during the preceding 24 h, was haemodynamically stable, had documentation in their chart of significant improvement of local symptoms, and was not oliguric (except for chronic oliguria in patients with end-stage renal disease).

Patients in the BL group were older (79.2 versus 71.6 years, P = 0.073) and had a significantly higher median CCI (6 versus 4, P < 0.001), a higher prevalence of advanced renal failure (41.7% versus 0%, P < 0.001), higher baseline creatinine levels (median 1.8 versus 1.0 mg/dL, P < 0.001), a higher CRP level on admission (183 versus 140.4 mg/L, P < 0.001) and higher Pitt bacteraemia score (median 2 versus 0, P < 0.001). There were no differences in the prevalence of urinary obstruction, urological interventions (including catheter placement) or time to appropriate antibiotic administration.

Gram-negative bacteria isolated from blood and urine cultures are shown in Table 2. There was no difference in the distribution of different species between groups. Overall, 34% of the isolates were ESBL-PE. Out of the patients treated with AGs, the isolates’ susceptibility to AGs were as follows: gentamicin MIC₅₀ ≤ 1 mg/L, MIC₉₀ ≥ 16 mg/L; amikacin MIC₅₀ ≤ 2 mg/L and MIC₉₀ = 4 mg/L. All patients were treated with an agent to which the isolate was susceptible.

Table 2.

Bacteria isolated from blood culture

	BL group	AG group	P value	Ceftriaxone resistant (% of species)
	n (%)	n (%)		n (%)
E. coli	69 (71.9)	26 (68.4)	0.691	26 (27.4)
Klebsiella pneumoniae	13 (13.5)	7 (18.4)	0.475	14 (70)
Proteus mirabilis	11 (11.4)	2 (5.3)	0.275	5 (38.5)
Enterobacter spp.	3 (3.1)	1 (2.6)	1	0 (0)
Citrobacter koseri	0 (0)	1 (2.6)	0.283	0 (0)
Serratia marcescens	0 (0)	1 (2.6)	0.283	0 (0)

Using univariate analysis, the primary outcome of clinical improvement within 72 h occurred in 55/96 (57%) of patients treated with a BL and 25/38 (66%) of patients treated with an AG (P = 0.335). Clinical improvement by the end of the admission period was observed in 84/96 (88%) and 36/38 (95%) of patients in the BL and AG groups, respectively. Hospital stay was shorter for the AG group (4.8 versus 6.5 days, P = 0.01). There were fewer in-hospital deaths in the AG group (5.3% versus 12.5%, P = 0.221) and fewer relapses (5.3% versus 12.5%, P = 0.221), but the differences were not statistically significant. There was no statistically significant difference between groups in time to clinical improvement.

The ratio of serum creatinine on Day 3 to Day 1 of treatment was lower in the BL group compared with the AG group (median 0.78 and 0.86, P = 0.012). A 50% rise in serum creatinine by Day 3 was seen in 2/91 (2.2%) and 3/35 (8.6%) of patients in the BL and AG groups, respectively (P = 0.130).

The results of a multivariate analysis are shown in Table 3. A logistic regression model was fitted using clinical improvement after 72 h or clinical improvement by the end of admission. In both analyses, the empirical antibiotic treatment (BL or AG) was not significantly associated with clinical outcome, with ORs of 1.52 (95% CI 0.54–4.31, P = 0.428) and 1.07 (95% CI 0.31–3.51, P = 0.910), respectively.

Table 3.

Predictors of clinical improvement using multivariate analysis

Outcome	Parameter	OR (95% CI)	P value
Clinical improvement after 72 h	Age	1.01 (0.98–1.05)	0.496
	Male sex	1.17 (0.51–2.78)	0.714
	CCI	0.93 (0.75–1.15)	0.482
	Pitt bacteraemia score	0.85 (0.7–1.04)	0.120
	Creatinine at presentation	0.629 (0.34–1.10)	0.117
	CRP at presentation	1 (0.99–1.00)	0.105
	Any urinary blockage	1.1 (0.41–3.04)	0.854
	Any urological source control	0.91 (0.35–2.39)	0.852
	Time to appropriate antibiotic treatment	0.73 (0.41–1.22)	0.237
	BL versus AG	1.52 (0.54–4.31)	0.428
Clinical improvement at end of admission	Age	0.94 (0.86–1.02)	0.472
	Male sex	0.72 (0.19–2.78)	0.631
	CCI	0.89 (0.66–1.23)	0.472
	Pitt bacteraemia score	0.91 (0.69–1.23)	0.510
	Creatinine at presentation	1.19 (0.55–2.74)	0.666
	CRP at presentation	0.99 (0.98–0.998)	0.014
	Any urinary blockage	0.65 (0.14–3.1)	0.579
	Any urological source control	1.95 (0.43–10.37)	0.402
	Time to appropriate antibiotic treatment	1.57 (0.65–5.7)	0.397
	BL versus AG	1.15 (0.14–7.1)	0.888

After propensity score weighting, baseline differences between groups were considerably smaller but not eliminated (Figure S1, available as Supplementary data at JAC-AMR Online). On weighted logistic regression analysis (Table 4), age, CCI, Pitt bacteraemia score, baseline serum creatinine and CRP were associated with the primary outcome, whereas antimicrobial treatment (BL versus AG) was not (OR 1.61, 95% CI 0.87–3.01, P = 0.130).

Table 4.

Predictors of clinical improvement after 72 h after propensity score weighting

Parameter	OR (95% CI)	P value
Age	1.04 (1.02–1.07)	<0.01
CCI	0.77 (0.65–0.90)	<0.01
Pitt bacteraemia score	0.84 (0.72–0.98)	0.023
Creatinine at presentation	0.62 (0.39–0.98)	0.045
CRP at presentation	1.00 (0.99–1.00)	0.121
BL versus AG	1.61 (0.87–3.01)	0.130

In light of the institutional guidelines recommendation to avoid AGs in patients with an estimated GFR of less than 30 mL/min, a sensitivity analysis was performed using patients with a creatinine level of ≤1.5 mg/dL on admission. Included were 32 and 30 patients in the AG and BL groups, respectively. No statistically significant differences were noted in the primary outcome, with an OR of BL for clinical improvement at 72 h of 1.24 (CI 0.28–6.00, P = 0.777). A 50% rise in serum creatinine by Day 3 was seen in 1/28 (3.6%) and 1/28 (3.6%) of patients in the BL and AG groups, respectively (P = 0.150). Another sensitivity analysis was performed to check if our findings were dependent on specific bacterial species. Using only patients with Escherichia coli infections (71% of all cases), no primary outcome difference was shown between the two treatment groups (OR for BL 1.18, 95% CI 0.3–4.44, P = 0.8).

Discussion

We present a retrospective analysis comparing the outcome of empirical AG or BL antibiotic treatment for Gram-negative bacteraemia of urinary origin in a single teaching hospital. The selection of empirical treatment during the first 72 h is important since early administration of appropriate antibiotics in this time frame is a major determinant of patient outcome.¹⁰ In our cohort, the rate of clinical improvement at 72 h when an empirical AG was used was not different from that of patients treated with a BL (66% and 57%, respectively), as for the rate of improvement at the end of admission (95% and 88%, respectively). These findings remained unchanged after multivariate analysis aimed at correcting for age, baseline differences in comorbidities, renal function and disease severity on presentation. Among secondary outcomes that were evaluated, AG treated patients had a statistically significant shorter hospital stay, although this can be attributed to their better baseline characteristics. As for the safety of AG treatment, renal function improved after 3 days of treatment in both groups. Improvement was slightly more pronounced in patients in the BL group, as reflected by lower Day 3/Day 1 serum creatinine ratio (0.778 versus 0.895, P = 0.012). This may represent a selection bias, since institutional protocols discourage the use of AGs in patients with significant renal failure, who are therefore more likely to be treated with BLs. As, in many cases, patients present with pre-renal acute kidney injury, those patients are likely to have improvement in their renal function with antibiotic and supportive care by Day 3. An increase in creatinine of 50% or more by Day 3 was seen more commonly in patients receiving AGs, but this difference was not statistically significant. When limiting the analysis to patients without significant renal failure on presentation, the nephrotoxicity rate was similar between groups (3.6%).

Antibiotic resistance is a growing global threat and is fuelled by the prescribing of broad-spectrum antibiotics. Among Gram-negative bacteria, the prevalence of ESBL-PE is rising worldwide, accounting for >40% of isolates in highly endemic areas such as Asia, Latin America and the Middle East.¹¹ Indeed, the prevalence in North America and Northern Europe is much lower (usually less than 10% of isolates),^11,12 but it has doubled between 2009 and 2014.¹³ Quinolone resistance is also prevalent, with more than 1/3 of E. coli isolates from inpatients being resistant.¹⁴ In Israel, a high prevalence of quinolone resistance in ESBL-PE has also been reported in young healthy outpatients (11.1% and 5.9%, respectively).¹⁵ These resistance patterns make third-generation cephalosporins and quinolones a questionable choice for empirical treatment in patients with urosepsis. Upfront use of broader spectrum BLs such as carbapenems and novel BL-β-lactamase inhibitors (BLBLIs) carries higher costs and may further fuel resistance, increasing the prevalence of carbapenemase-producing Enterobacterales. This reality makes AGs an appealing choice for empirical treatment in cUTI.

AGs were extensively used for UTI from the 1940s, but the introduction of BLs and quinolones, with their improved safety profile, led to the replacement of AGs as first-line empirical and definitive treatments. However, changes made to the use of AGs, including a shift to once-daily dosing, and routine monitoring of trough plasma drug concentration, have resulted in a lower risk of toxicity. Furthermore, a short duration of treatment, as indicated for UTI, is associated with lower rates of adverse effects.^16–18 Although nephrotoxicity remains a significant risk with AG use, it is usually mild and temporary, especially with short courses.¹⁹ The nephrotoxic risk in patients with sepsis seems to be much more related to the haemodynamic state and comorbidities than to AG use.²⁰ A meta-analysis of randomized controlled trials (RCTs) comparing AGs to other agents including BLs and quinolones, showed AGs to be non-inferior for clinical cure and survival. Rates of microbiological cure were lower with AGs at 7 days after treatment, but not at 30 days. Importantly, in most studies that evaluated AGs, multiple daily dosing was used, and only 17% of patients had sepsis.²¹ In a recent RCT, the novel AG plazomicin was non-inferior to meropenem at the end of treatment for cUTI, and even superior in long-term outcome.²² In a retrospective study, non-carbapenem antibiotics, including AGs, were as good as carbapenems for the treatment of pyelonephritis with ESBL-PE.²³ A particular advantage of AGs is their low penetration into the gastrointestinal lumen, resulting in less perturbation of the colonic microbiota. This advantage was used for treatment of recurrent UTI in patients with a history of Clostridioides difficile colitis, with no recurrence and no apparent effect on faecal microbiota examined.²⁴

Elbaz et al. described a single-centre study of patients treated for UTI after an institutional policy had changed to recommend AGs as first-line treatment.²⁵ The study compared 715 patients treated with AGs with 1311 patients treated with comparators, mostly BLs. Treatment with an AG was twice as likely to be empirically appropriate and had a 22% reduced risk of mortality. Although appropriateness of the treatment can be responsible for this survival benefit, the effect remained for cases of ESBL-PE infection, which are the main reason for the inappropriateness of BLs. Zohar et al. retrospectively compared patients with ESBL-PE bacteraemia of urinary origin who received AGs with those who received either piperacillin/tazobactam or carbapenems.²⁶ AGs were non-inferior to BLs regarding mortality, and were not associated with more nephrotoxicity.

Our study shows the comparable efficacy of AGs for the treatment of patients with bacteraemic urosepsis. The non-inferiority of AGs was shown in multiple important outcomes including clinical improvement at different timepoints, survival and length of stay. Including only patients with bacteraemia in this study enabled a more definitive diagnosis of UTI and prevented inclusion of patients with sepsis of a non-urinary origin who had bacteriuria. By extrapolation, the finding that AGs are non-inferior for patients with urosepsis supports their use in milder forms of UTI.

This study has some important limitations. The confidence of the main finding (lack of major difference between groups) is limited by the small number of included patients, and therefore minor differences might exist. The same is true for nephrotoxicity. Due to its retrospective observational design, the compared populations of treated patients were not identical, and BL-treated patients tended to have more comorbidities, be more severely ill and have reduced renal function. We were able to partially correct for differences in baseline characteristics using propensity score weighting; however, some differences remained, and other unidentified differences could influence the outcomes. We examined only the effect of antibiotics that were appropriate according to in vitro testing, ignoring empirical non-appropriate antibiotics given first, which might have influenced the outcome. We have only estimated nephrotoxicity in our safety analysis, and did not examine ototoxicity, which is another important adverse effect of AGs.

In conclusion, this single-centre study, performed in a community with a remarkably high prevalence of ESBL-PE, shows that AGs are an effective and safe treatment for cUTI in bacteraemic patients. The rapidly growing epidemic of ESBL-PE and carbapenem-resistant Gram-negative bacteria underscores the need for re-consideration for AG use as first-line agents in patients with UTI.

Funding

This study was carried out as part of our routine work.

Transparency declarations

The authors declare they have no financial interests.

Author contributions

Study conceptualization: T.B.N.; data curation: R.P.; data analysis: R.P., A.C.; laboratory data analysis: N.S.; writing—original draft preparation: T.B.N.; writing—review and editing: A.C., R.P., N.S.; oversight: T.B.N.

Data availability

De-identified original data can be shared by the corresponding author upon reasonable request.

Supplementary data

Figure S1 is available as Supplementary data at JAC-AMR  Online.