Less Is More: Combination Antibiotic Therapy for the Treatment of Gram-Negative Bacteremia in Pediatric Patients

Abstract

IMPORTANCE

Definitive combination antibiotic therapy with a β-lactam and an aminoglycoside for the treatment of gram-negative bacteremia is commonly prescribed in pediatric patients; however, its efficacy and toxicity relative to β-lactam monotherapy are unknown.

OBJECTIVE

To determine whether definitive combination antibiotic therapy affects mortality and nephrotoxicity in pediatric patients with gram-negative bacteremia.

DESIGN, SETTING, AND PARTICIPANTS

Retrospective cohort study including pediatric patients (aged ≤18 years) with gram-negative bacteremia hospitalized at the Johns Hopkins Children’s Center between 2002 and 2011.

MAIN OUTCOMES AND MEASURES

Outcomes included 30-day mortality and nephrotoxicity classified according to the pediatric RIFLE (risk for renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end-stage renal disease) criteria. To account for nonrandom assignment of combination therapy, propensity score weighting was combined with multivariable logistic regression to estimate the effect of combination therapy on mortality and nephrotoxicity.

RESULTS

Of the 879 eligible pediatric patients with bacteremia, 537 (61.1%) received combination therapy. After propensity score adjustment, baseline demographic and clinical characteristics between the groups were well balanced. There was no association between combination therapy and 30-day mortality (odds ratio, 0.98; 95% CI, 0.93–1.02; P = .27). There were 170 patients (19.3%) with evidence of acute kidney injury, including 135 (25.1%) and 35 (10.2%) in the combination therapy and monotherapy arms, respectively. Patients receiving combination therapy had approximately twice the odds of nephrotoxicity compared with those receiving monotherapy (odds ratio, 2.15; 95% CI, 2.09–2.21).

CONCLUSIONS AND RELEVANCE

The use of β-lactam monotherapy for gram-negative bacteremia in pediatric patients reduces subsequent nephrotoxicity without compromising survival.

There is considerable debate about the role of β-lactam monotherapy compared with combination therapy including β-lactams and aminoglycosides for the definitive treatment of gram-negative bacteremia in pediatric patients.^1–3 Although combination antibiotic therapy may be appropriate to expand empirical antimicrobial coverage in ill pediatric patients, the continued benefit of combination therapy after susceptibility data are available is uncertain. Several studies in adults,^4–11 including meta-analyses of randomized clinical trials, have demonstrated no difference in clinical outcomes between combination therapy and monotherapy as culture-directed therapy for gram-negative bacteremia. To date, this issue has not been formally evaluated in the pediatric population, and combination antibiotic therapy continues to be liberally prescribed, particularly for organisms perceived to be especially virulent, such as Pseudomonas aeruginosa.^1–3

It is concerning that the addition of aminoglycoside therapy may result in acute renal injury in adults.^{5–8,12–15} Aminoglycosides can accumulate in the renal cortex and cause acute nephrotoxicity, which may not always be reversible.^16–19 A substantial advantage can be gained by simplifying an antibiotic regimen with a single drug, provided that the agent is effective and well tolerated.

We hypothesized that combination antibiotic therapy for gram-negative bacteremia increases the risk of acute renal injury, without improving survival compared with monotherapy in the pediatric population. We conducted a retrospective cohort study in consecutive pediatric patients with gram-negative bacteremia during a 10-year period to determine whether the use of combination therapy as definitive therapy for gram-negative bacteremia affects subsequent survival and nephrotoxicity compared with monotherapy.

Methods

Setting and Participants

We identified pediatric patients (aged ≤18 years) admitted to the Johns Hopkins Children’s Center in Baltimore, Maryland, from January 1, 2002, through December 31, 2011, with Enterobacteriaceae, Pseudomonas species, or Acinetobacter species cultured from a sample of blood with clinical signs and symptoms suggestive of infection, given age-based normal values (temperature >38°C or <36°C, leukopenia, leukocytosis, apnea, bradycardia, tachycardia, or hypotension).²⁰ Patients were excluded if they (1) had polymicrobial bacteremia, (2) did not receive appropriate gram-negative therapy (to which the organism ultimately recovered was susceptible) within 24 hours after obtaining their first positive blood culture result, (3) received an antibiotic regimen other than a β-lactam with or without the addition of an aminoglycoside, or (4) died within 48 hours after obtaining their first positive blood culture result. Only the first episode of bacteremia due to the same organism per patient was included in the analysis.

Data Collection

Laboratory databases were queried to identify all blood cultures from which gram-negative organisms were isolated during the study period. Characteristics of pediatric patients with gram-negative bacteremia were extracted from medical records. The primary exposure was use of combination therapy. Outcomes of interest were mortality and nephrotoxicity. Pertinent additional data included demographic characteristics, lowest absolute neutrophil count within 48 hours after the first positive blood culture result, and other body sites where the relevant organisms were recovered. The highest Pediatric Risk of Mortality (PRISM) III score, admission to the intensive care unit (ICU), vasopressor requirement, and need for mechanical ventilation were recorded as measures of illness severity on the day the first positive blood culture result was obtained (within 24 hours). Additional data collected as hypothesized confounders included failure to remove all central venous lines within 72 hours after obtaining the first positive blood culture result and time at risk (duration [in days] from hospital admission to the first positive blood culture). Because of existing data suggesting that Pseudomonas bacteremia is associated with higher mortality than other gram-negative organisms causing bacteremia, the presence of Pseudomonas bacteremia was also treated as a potential confounder.^21,22 The use of intravenous contrast material or more than 48 hours of additional nephrotoxic drug treatment (amphotericin B, colistimethate sodium, cyclosporine, acyclovir, vancomycin hydrochloride, cidofovir, cyclophosphamide, methotrexate sodium, or cisplatin) were considered as dichotomous variables because they also may be associated with the development of nephrotoxicity. This study was approved by the Johns Hopkins University School of Medicine Institutional Review Board, with a waiver of the requirement for informed consent.

Definitions

Antimicrobial therapy administered before final susceptibility results were available was considered empirical, and therapy administered after these results was considered definitive. The term combination therapy was used when a β-lactam and an aminoglycoside were prescribed as definitive therapy for at least 48 hours after antibiotic susceptibilities were finalized. Monotherapy referred to the use of a β-lactam alone as definitive therapy. Mortality was defined as death from any cause within 30 days from the first positive blood culture result. Among the patients discharged less than 30 days after initiation of antibiotics, more than 95% returned for additional clinic visits or had future hospitalizations, which allowed us to determine that they were alive at 30 days.

Nephrotoxicity was categorized as a dichotomous variable according to the pediatric RIFLE (risk for renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end-stage renal disease) criteria,^23,24 which are used to stratify pediatric patients based on changes in serum creatinine levels from baseline and/or changes in urine output. We limited classification to risk, injury, and failure according to the maximum RIFLE class reached up to 72 hours after discontinuation of antibiotics. Baseline serum creatinine levels were determined by using the lowest level measured within 72 hours after obtaining the first positive blood culture result. Risk was defined as an increase in serum creatinine to 1.5 times baseline and/or urine output less than 0.5 mL/kg/h for at least 6 hours; injury was defined as an increase in serum creatinine to at least twice baseline and/or urine output less than 0.5 mL/kg/h for at least 12 hours; and failure was defined as an increase in serum creatinine to at least 3 times baseline and/or urine output less than 0.3 mL/kg/h for at least 24 hours or anuria for at least 12 hours.

Statistical Analysis

Demographic and clinical characteristics were summarized as percentages for categorical variables and means and standard deviations for continuous variables. Year was regressed on the proportion of patients who received combination therapy and separately on the annual proportion of patients who died to evaluate trends during the study period. Because it was anticipated that the distribution of baseline characteristics would differ substantially between the patients who received combination antibiotic therapy and those who received monotherapy, propensity scores were created to establish each patient’s probability of receiving combination therapy as definitive therapy.

Propensity scores were estimated for each patient by using boosted classification trees.²⁵ Classification trees were implemented with 10 000 iterations and an iteration stopping point that minimized the mean of the Kolmogorov-Smirnov test statistics. The covariates used to generate the propensity scores were age, number of preexisting medical conditions, time at risk, ICU admission, use of vasopressors, absolute neutrophil count no more than 100 cells/μL, mechanical ventilation, presence of a central venous line 72 hours after obtaining the first blood culture result, Pseudomonas bacteremia, and underlying cancer.

The average treatment effect of the treated, which compares the outcomes in treated patients (receiving definitive combination therapy) with those in patients with similar pre-treatment characteristics who received monotherapy, was chosen as the estimand.^26,27 Patients receiving monotherapy were weighted by their odds of receiving combination therapy, trimmed at 0.10 to create balanced covariate distributions.^28,29 Covariate balance after weighting was assessed by using the absolute standardized mean difference, the Kolmogorov-Smirnov test statistic, the t test for continuous variables, and Pearson χ² test for categorical values. The absolute standardized mean difference was 0.20 or less, and the Kolmogorov-Smirnov test statistic was less than 0.10 for all variables used in the development of propensity scores after weighting.

Unadjusted odds ratios (ORs) were estimated for the association between the use of combination therapy and mortality and the use of combination therapy and nephrotoxicity in unweighted univariable analysis. All covariates with P < .20 in the unadjusted model were entered into a doubly robust, adjusted model incorporating weights and potential confounders. Potential confounders were identified as variables producing a risk estimate that differed by more than 10% from the crude unadjusted risk estimate when included in the final model. For all statistical tests, differences were considered statistically significant at P < .05 (2-sided). Data were analyzed with Stata software (version 11.1; StataCorp) and the twang package for the R programming language (version 2.14.1; R Development Core Team).³⁰

Results

Study Population

There were 1166 pediatric patients with gram-negative bacteremia treated between January 1, 2002, and December 31, 2011, of whom 879 met eligibility criteria (Figure). Klebsiella pneumoniae was the most common organism identified, followed by P aeruginosa and Escherichia coli (Table 1). The combination therapy group included 537 patients (61.1%), and the monotherapy group, 342 (38.9%) (Table 2). There was no statistically significant change in the proportion of patients receiving combination therapy during the study period (P = .12). Piperacillin-tazobactam was the most commonly prescribed β-lactam (37.3%), followed by ceftriaxone (30.8%) and cefepime hydrochloride (15.0%).

Figure.

Study Design

Design of study in pediatric patients receiving combination antibiotic therapy vs monotherapy for definitive treatment of gram-negative bacteremia (due to Enterobacteriaceae, Pseudomonas species, or Acinetobacter species) between 2002 and 2011. Some patients met more than 1 exclusion criterion.

Table 1.

Microbiological Findings in 1166 Pediatric Patients Hospitalized With Gram-Negative Bacteremia From 2002 Through 2011

Organism	No. (%)
Acinetobacter baumannii	69 (5.9)
Escherichia coli	194 (16.6)
Citrobacter species	44 (3.8)
Citrobacter freundii	41 (3.5)
Citrobacter koseri	3 (0.3)
Enterobacter species	207 (17.8)
Enterobacter aerogenes	22 (1.9)
Enterobacter agglomerans	11 (0.9)
Enterobacter asburiae	6 (0.5)
Enterobactercloacae	168 (14.4)
Klebsiella species	321 (27.5)
Klebsiella oxytoca	53 (4.5)
Klebsiella ozaenae	18 (1.5)
Klebsiella pneumoniae	250 (21.4)
Proteus mirabilis	5 (0.4)
Pseudomonas species	257 (22.0)
Pseudomonas aeruginosa	239 (20.5)
Pseudomonas fluorescens	15 (1.2)
Pseudomonas putida	3 (0.3)
Serratia marcescens	69 (5.9)

Table 2.

Demographic and Clinical Characteristics of 879 Pediatric Patients Receiving Combination Therapy or Monotherapy for Gram-Negative Bacteremia, Before and After Propensity Score Adjustment^a

Characteristic	Combination Therapy (n = 537)	Monotherapy, Unweighted (n = 342)	P Value, Combination vs Unweightedb	Monotherapy, Weightedc	P Value, Combination vs Weightedb
Age, mean (SD), y	5.7 (6.4)	6.4 (7.0)	.09	5.7 (6.5)	0.99
Female sex	39.9	43.6	.28	44.8	0.18
Preexisting medical conditions
Prematurity (≤33wk gestation)	16.2	12.0	.08	12.3	.14
Hematological	3.9	8.5	.01	11.3	.07
Neuromuscular	6.1	9.4	.09	10.0	.08
Cardiovascular	8.2	5.8	.18	7.0	.55
Respiratory	6.7	5.6	.49	8.0	.66
Gastrointestinal	30.5	18.1	<.001	21.0	.01
Renal	8.4	9.4	.62	10.0	.58
Immunocompromisedd	16.0	22.2	.02	20.3	.16
Cancer	21.8	38.9	<.001	26.3	.20
Genetic or metabolic	10.4	7.0	.08	9.0	.56
Second- or third-degree burns	1.0	1.0	.74	1.4	.58
No. of preexisting conditions, mean (SD)	1.3 (0.6)	1.2 (0.6)	.06	1.3 (0.6)	.19
Highest PRISM III score, mean (SD)	7.6 (8.8)	7.1 (7.6)	.37	7.7 (8.4)	.99
ICU admission	46.7	35.7	.001	45.0	.57
Vasopressors	18.8	13.7	.04	19.2	.96
Mechanical ventilation	29.1	15.5	<.001	24.6	.19
Time at risk, mean (SD), de	17.9 (33.4)	16.8 (35.2)	.09	19.4 (39.5)	.65
Absolute neutrophil count ≤100 cells/μL	13.1	10.2	.01	12.2	.35
Central line in place 72 h after first positive blood culture result	58.1	49.1	.01	54.6	.45
Pseudomonas species bacteremia	29.1	16.1	<.001	19.3	.01
Other body sites where organism was recovered
Urine	12.7	7.3	.01	6.5	.003
Pleural or bronchoalveolar lavage fluid	4.5	5.3	.60	2.5	.27
Bone or joint specimens	1.0	2.0	.13	3.5	.08
Intra-abdominal fluid	1.0	1.8	.32	1.6	.61
Cerebrospinal fluid	1.7	1.0	.12	1.3	.60

Abbreviations: ICU, intensive care unit; PRISM, Pediatric Risk of Mortality.

^aCombination therapy included a β-lactam and an aminoglycoside; monotherapy, only a β-lactam. Data represent percentages unless otherwise indicated.

^bP value for t statistic for continuous variables and Pearson χ² statistic for categorical variables.

^cThe effective sample size was 233; this value captures the increase in sampling variance created by weighted means and gives an estimate of the number of patients receiving monotherapy who are comparable to those receiving combination therapy.

^dImmunocompromised patients include those receiving corticosteroid therapy (≥2 mg/kg for ≥14 d), immunomodulator therapy, hematopoietic stem cell transplant ≥1 y before onset of bacteremia, solid organ transplant, or cancer chemotherapy ≥6 mo before onset of bacteremia; those with congenital immunodeficiency; and those positive for human immunodeficiency virus (CD4⁺ T-cell count, <200 cells/mL).

^eTime at risk was defined as the number of days from hospital admission until the first positive blood culture result.

Baseline Characteristics

There were several clinically relevant differences between the treatment groups. Compared with patients receiving monotherapy (unweighted analysis), those receiving combination therapy were more likely to require ICU admission, vasopressors, or mechanical ventilation; to have a central venous line that remained in place for more than 72 hours after obtaining the first positive blood culture result; to have an underlying cancer; and to have Pseudomonas bacteremia. Patients with urosepsis as their source of bacteremia were much more likely to receive monotherapy (P = .01). A central venous line was identified as the likely source of bacteremia in 74.1% of patients. Most patients (92.2%) meeting inclusion criteria had at least 1 underlying medical condition; only 33 (3.8%) had no underlying medical conditions at the time gram-negative organisms were cultured from the samples of blood. After patients receiving monotherapy were weighted—to create a pseudo-population in which patients similar to those who received combination therapy were up-weighted and patients whose demographic and clinical characteristics made them unlikely to receive combination therapy were down-weighted— baseline demographic and clinical characteristics between the groups were well balanced. A notable exception was that patients with Pseudomonas bacteremia remained more likely to receive combination therapy (P = .01).

Mortality

There were a total of 128 deaths (11.0%) within 30 days after the onset of bacteremia in the total sample of 1166 pediatric patients. Half of them occurred in the first 48 hours after onset, and these patients were excluded from further analysis. The proportion of deaths in pediatric patients with gram-negative bacteremia remained consistent throughout the study period (P = .72);there were 41 deaths (7.6%)in the combination therapy and 23 (6.7%) in the monotherapy group (P = .61). In the weighted multivariable regression model adjusting for presence of a central venous line more than 72 hours after obtaining the first positive blood culture result, absolute neutrophil count of 100 cells/μL or less, ICU admission, vasopressor requirement, Pseudomonas bacteremia, mechanical ventilation, and PRISM III score, there was no association between combination therapy and mortality (OR, 0.98; 95% CI, 0.93– 1.02) (Table 3). More important, failure to remove central venous lines within 72 hours after detection of bacteremia was associated with more than 6 times the odds of mortality in an unadjusted model (OR, 6.46; 95% CI, 3.04–13.71); this association was attenuated, although still significant, after controlling for relevant confounders (OR, 2.11; 95% CI, 2.07–2.15).

Table 3.

Mortality Before and After Propensity Score Weighting Combined With Multivariable Logistic Analysis Comparing Combination Therapy and Monotherapy in 879 Pediatric Patients With Gram-Negative Bacteremia

	Unadjusted Univariable Analysis		Adjusted, Weighted Multivariable Analysisa
Characteristic	OR(95% CI)	P Value	OR(95% CI)	P Value
Age	1.00 (0.96–1.03)	.83	…	…
Combination therapy (β-lactam plus aminoglycoside)	1.14 (0.68–1.95)	.61	0.98 (0.93–1.02)	.27
Central line in place 72 h after first positive blood culture result	6.46 (3.04–13.71)	<.001	2.11 (2.07–2.15)	<.001
Cancer	1.07 (0.60–1.84)	.82	…	…
Absolute neutrophil count ≤100cells/μL	1.95 (1.14–3.37)	.03	1.05 (1.00–1.11)	.14
ICU admission	5.44 (2.96–10.00)	<.001	0.99 (0.94–1.04)	.63
Vasopressor requirement	8.07 (4.74–13.75)	<.001	1.02 (0.94–1.10)	.70
Pseudomonas bacteremia	1.34 (0.88–2.04)	.18	1.02 (0.98–1.05)	.37
Mechanical ventilation	8.67 (4.98–15.10)	<.001	1.03 (0.97–1.09)	.37
PRISM III score	1.12 (1.10–1.15)	<.001	1.01 (1.01–1.02)	<.001

Abbreviations: ICU, intensive care unit; OR, odds ratio; PRISM, Pediatric Risk of Mortality.

^aThe adjusted, weighted model for mortality includes failure to remove a central line within 72 h after the first positive blood culture result, absolute neutrophil count ≤100 cells/μL, ICU admission, vasopressor requirement, mechanical ventilation, highest PRISM III score on the day of the first positive blood culture result (within 24 hours), and Pseudomonas bacteremia.

Nephrotoxicity

Nephrotoxicity developed in 170 pediatric patients (19.3%) receiving antimicrobial therapy (Table 4). Acute renal toxicity developed in 135 (25.1%) of those receiving combination therapy compared with 35 (10.2%) of those receiving monotherapy. The median time to nephrotoxicity was 5.6 days after aminoglycoside therapy was initiated. Compared with those receiving monotherapy, patients receiving combination therapy had more than twice the odds of nephrotoxicity (OR, 2.95; 95% CI, 1.97–4.41); this relationship persisted after adjustment for the receipt of additional nephrotoxic agents, PRISM III score, and age (OR, 2.15; 95% CI, 2.09–2.21). Of patients receiving aminoglycoside therapy who had nephrotoxicity, 68.8%, 27.4%, and 3.7% were in the risk, injury, and failure categories, respectively, according to pediatric RIFLE criteria.

Table 4.

Mortality and Nephrotoxicity Before and After Propensity Score Weighting Combined With Multivariable Logistic Analysis Comparing Combination Therapy and Monotherapy in 879 Pediatric Patients With Gram-Negative Bacteremia

Characteristic	No. (%) of Events by Treatment Group		Unadjusted OR (95% CI)	P Value	Adjusted, Weighted OR(95% CI)	P Value
	Combination Therapy (n = 537)	Monotherapy (n = 342)
Mortalitya	41 (7.6)	23 (6.7)	1.14 (0.68–1.95)	.61	0.98 (0.93–1.02)	.27
Nephrotoxicityb	135 (25.1)	35 (10.2)	2.95 (1.99–4.45)	<.001	2.15 (2.09–2.21)	<.001
Riskc	93 (68.8)	20 (57.1)	…	…	…	…
Injury	37 (27.4)	14 (40.0)	…	…	…	…
Failure	5 (3.7)	1 (2.9)	…	…	…	…

Abbreviation: OR, odds ratio.

^aThe adjusted, weighted model for mortality is explained in footnote to Table 3.

^bThe adjusted model for nephrotoxicity includes age, Pediatric Risk of Mortality III score, and the use of nephrotoxins (intravenous contrast material or >48 h of any of the following medications as a dichotomous variable: amphotericin B, colistimethate sodium, cyclosporine, acyclovir, vancomycin hydrochloride, cidofovir, cyclophosphamide, methotrexate sodium, or cisplatin).

^cPercentages represent number of patients for the nephrotoxicity category divided by the number of all patients in the treatment group with nephrotoxicity according to the pediatric RIFLE (risk for renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end-stage renal disease) criteria. Statistical analysis was not performed for risk, injury, or failure because of the small sample sizes.

Discussion

Our study demonstrates that the risk of mortality is similar in pediatric patients with gram-negative bacteremia treated with β-lactam monotherapy and those treated with combination (β-lactam and aminoglycoside) therapy. However, the use of combination therapy may predispose pediatric patients to acute renal injury. In our cohort, 61.1% of patients received both a β-lactam and an aminoglycoside for definitive treatment of gram-negative bacteremia, even though the organism was susceptible to the β-lactam prescribed. “Two drugs are better than one” continues to be a common belief in pediatrics, although existing clinical data from adult experiences do not support this dogma.^1,2 At least 8 meta-analyses,^{4–8,10,11,31} including more than 70 randomized clinical trials, have addressed the issue of definitive combination antibiotic therapy in adults with gram-negative bacteremia. In all these meta-analyses, mortality did not differ between treatment groups, and in the 6 meta-analyses evaluating nephrotoxicity, the rate of acute renal injury was significantly higher in the combination therapy arm. To our knowledge, ours is the first study addressing the comparative advantages of combination therapy vs mono-therapy in pediatric patients.

Ample evidence suggests that appropriate empirical therapy leads to lower mortality and improved clinical outcomes in patients with gram-negative bacteremia.^32–40 In the era of increasingly resistant organisms, “appropriate” empirical therapy often includes the addition of an aminoglycoside to ensure that the organism recovered will be susceptible to at least 1 antimicrobial agent prescribed. Once organism identification and susceptibility testing are complete, the antibiotic regimen should be fine-tuned to balance providing optimal therapy with limiting unnecessary antibiotic exposure.⁴¹

In our cohort, pediatric patients who received definitive combination therapy had twice the odds of nephrotoxicity, defined using the pediatric RIFLE criteria. This is in contrast to the notion—frequently expressed in pediatrics—that aminoglycoside use rarely causes nephrotoxicity in children.² There is growing evidence that nephrotoxicity, which may not always be reversible, is associated with aminoglycoside use in the first few years of life.^16–19 This possible association is especially concerning because children with gram-negative bacteremia often require additional nephrotoxic agents for sepsis management, potentially worsening renal injury caused by aminoglycosides. Other disadvantages of adding an aminoglycoside to a gram-negative treatment regimen, which we did not address, include ototoxicity, the need for frequent catheter access (placing patients at risk for subsequent infections), the need for therapeutic monitoring of drug levels, and additional drug acquisition, preparation, and administration costs.⁴¹

Pseudomonas bacteremia has been shown to increase mortality compared with bacteremia due to most other gram-negative organisms.^21,22 Because of this concern, the pediatric literature recommends combination therapy for the treatment of Pseudomonas bacteremia.^1–3 In our cohort, there was no difference in Pseudomonas bacteremia–associated mortality between the combination therapy and monotherapy groups in an adjusted model. Several meta-analyses^{5–8,10,11,42} of randomized clinical trials in adult populations with Pseudomonas infections have included subgroup analyses evaluating the incremental benefit of aminoglycoside therapy, and all but 1 meta-analysis found no clinical advantage to combination therapy. The authors of 1 meta-analysis⁴² concluded that combination therapy should be used when P aeruginosa is recovered from the bloodstream; however, when patients receiving aminoglycoside monotherapy were excluded from this meta-analysis, clinical outcomes did not differ between the treatment arms.⁴³

In our cohort, patients with evidence of gram-negative sepsis whose central venous lines were not removed within 72 hours after obtaining the first positive blood culture result were more likely to die within 30 days of this culture than those whose central venous lines were removed. The importance of removing central lines in patients with septicemia cannot be overstated. Many studies in both the adult and the pediatric literature have demonstrated that retention of central lines is more likely to be associated with mortality than catheter removal in patients with candidemia,^44–48 Staphylococcus aureus bacteremia,^49–52 or gram-negative bacteremia.^53–56

Several limitations should be considered when interpreting our findings. First, although we demonstrated that the addition of an aminoglycoside to a β-lactam seems to increase nephrotoxicity, we monitored serum creatinine levels only until 72 hours after completion of antibiotic regimens. We cannot draw any conclusions regarding lingering effects of acute renal injury in pediatric patients receiving aminoglycoside therapy. It is unclear whether aminoglycoside-related nephrotoxicity in childhood contributes to “multifactorial” causes of chronic renal failure later in life. Further research into long-term complications related to aminoglycoside use in pediatric patients is necessary to address this issue.

Second, evaluating treatment effects from observational data can be problematic because prognostic factors may influence treatment decisions. We attempted to overcome this potential selection bias with the inclusion of propensity scores. The test of a good propensity score model is that it adequately balances confounders between exposure groups. Our model achieved this balance of measured demographic and clinical variables relatively well. However, propensity scores cannot balance unknown confounders. As in all observational studies, we cannot rule out residual confounding by un-measured variables related to disease severity; however, our final model demonstrated minimal residual confounding that would have been unlikely to change our estimate of the association between combination therapy and mortality.

Finally, for 74.1% of patients in our cohort, a central venous line was the most likely source of bacteremia. We did not have adequate power to determine whether there is a benefit to adding aminoglycosides in the presence of a deep-seated source of infection, such as intra-abdominal abscess or infections of the lungs or meninges. However, because aminoglycosides are known to poorly penetrate abscesses, bronchial secretions, and cerebrospinal fluid,^57–61 we hypothesize that adding an aminoglycoside is unlikely to improve clinical outcomes in patients with such involvement.

In conclusion, we have identified an important problem in pediatrics that can help change our paradigm for treating gram-negative bacteremia in pediatric patients. Combining our results with the large body of evidence supporting β-lactam monotherapy for gram-negative bacteremia in the adult literature, we believe that β-lactam monotherapy in pediatric patients with gram-negative bacteremia can decrease the development of nephrotoxicity without compromising clinical outcomes.