Data Sources  Online bibliographies were searched for articles related to serious bacterial infection and fever in infants. Reference lists from selected and review articles were also examined.

Study Selection  Studies that reported rates and types of bacterial infection in febrile outpatients younger than 3 months were included. Those performed outside North America, lacking results by age, or those that evaluated selected patient populations were excluded.

Data Extraction  Two authors independently reviewed the selected articles for inclusion and abstracted the data.

Data Synthesis  Fourteen studies, evaluating 5247 febrile outpatients, were included. The prevalences of L monocytogenes and enterococcal infections were 7.3 (binomial exact 95% confidence interval [CI], 3.5-13.3), 1.9 (95% CI, 0.6-4.4), and 5.6 (95% CI, 0.7-2.1) per 1000 febrile infants in the first, second, and third months of life, respectively. To cover 1 infant with serious bacterial infection caused by L monocytogenes and enterococcal infections, the numbers of febrile infants who would need ampicillin were estimated as 138 (95% CI, 76-288) in the first month, 527 (95% CI, 226-1621) in the second month, and 178 (95% CI, 50-1469) in the third month. Enterococcal infections occurred in all ages studied; there were no Listeria infections after 30 days of age.

Conclusion  The empirical use of ampicillin to cover febrile infants for L monocytogenes and enterococcal infections is most justifiable in the first month of life.

Ampicillin is commonly used in 2 regimens: for the empirical treatment of (1) neonatal sepsis (ampicillin plus an aminoglycoside) and (2) neonatal meningitis (ampicillin plus a third-generation cephalosporin). Its inclusion is generally suggested to improve coverage for L monocytogenes, and there is the additional potential benefit of coverage for infections caused by enterococcus. Neither of these organisms is susceptible to the cephalosporins. Either ampicillin plus an aminoglycoside or a third-generation cephalosporin alone provides excellent empirical coverage against the most common neonatal pathogens, group B streptococcus and Escherichia coli.³

Although enterococcal infections are generally believed to be acquired after birth, L monocytogenes is thought to be perinatally acquired. Thus, it is commonly assumed that an infant's risk of listeriosis diminishes with increased postnatal age. In 8 population-based studies^4-11 with adequate data for analysis, 340 (95%) of 357 confirmed cases of L monocytogenes occurred in the first 30 days of life.^4-11 However, it does not necessarily follow that febrile outpatient infants are more likely to have L monocytogenes if they present in the first month vs subsequent months, since their fever changes their risk of bacterial disease compared with the general population. Other perinatally acquired infections, such as group B streptococcus and E coli, may also decrease over time; thus, the proportion of febrile infants with L monocytogenes could theoretically be constant during the first few months of life.

We conducted a systematic literature review to estimate the prevalence of L monocytogenes and enterococcal infections among febrile infants in the first, second, and third months of life. We used this information to estimate the number of febrile infants who would need empirical ampicillin therapy to provide prompt, appropriate antibiotic coverage for 1 infant infected with these bacteria.

METHODS

Two bibliographic databases, MEDLINE and the Cochrane Controlled Trials Register, were searched in December 2000 for the following terms: (1) serious bacterial infection.tw (tw indicates textword) or *bacterial infection (* indicates explode of a medical subject heading) or *sepsis or *bacteremia and (2) *fever or fever.tw or febrile.tw and (3) limited to "newborn" or "infant." In addition, references from included studies and recent review articles were reviewed. Studies were included if they evaluated febrile infants younger than 3 months in any outpatient setting (emergency department, clinic, or physician's office) for the presence of bacterial infection. All authors' definitions of fever were accepted. At least 75% of study infants needed to have blood and CSF cultures performed to be analyzed for B/BM, and, in addition, 75% needed to have urine cultures performed to be analyzed for SBI. The studies had to report the prevalence of L monocytogenes or enterococcal infection according to patient age in either 4-week or 1-month intervals. Studies were excluded if they evaluated only a subset of febrile infants (eg, low-risk infants). If there were multiple studies from the same institution with overlapping enrollment periods, then the studies were selected to include the largest independent population of infants.

One of us (J.C.B.) evaluated study titles and abstracts and retrieved the full text of all studies that might include infants younger than 3 months. All studies that contained data on bacterial infections in infants up to 3 months of age were then blinded to author and journal for review by the second author (J.L.B.). Both authors independently determined which studies met inclusion and exclusion criteria. Disagreements were resolved by discussion, and consensus was achieved in the selection of articles for analysis. Attempts were made to contact study authors when small amounts of additional information would allow their studies to be included. Data were then independently abstracted by both reviewers.

Estimates were made for 3 categories of age in 1-month blocks. Prevalences between groups were compared using the Fisher exact test. Within each age group, we estimated the prevalence of Listeria and enterococcal infections by dividing the total number of these infections by the total number studied. We calculated exact 95% binomial confidence intervals. These may be thought of as intervals for fixed-effects estimates, which assume that the underlying prevalence of Listeria and enterococcal infections was the same in all studies.

We also used bias-corrected bootstrap methods to generate 95% confidence intervals.^12-13 First, we sampled within each study to account for within-study variation. Second, we sampled the studies themselves to account for between-study variation. These may be thought of as intervals for random-effects estimates, which allow for the possibility that the true prevalence of Listeria and enterococcal infections differed across studies. Analyses were done using Stata statistical software.¹⁴

Finally, we estimated how many febrile children must receive empirical therapy with ampicillin to provide prompt empirical treatment for 1 child with Listeria and enterococcal infections. This was simply the inverse of the prevalence estimates and may be thought of as the number needed to cover (NNC).

RESULTS

Ninety-two studies^{2, 32-122} were found to have a potentially relevant population. Attempts were made to contact 7 study authors^{33, 43, 50, 57, 71, 110, 118} to obtain additional data, with 5 responses; 2 of these authors^{33, 50} provided additional information.

Ultimately, 14 studies met criteria for inclusion (Table 1).^32-45 Nine studies were prospective, and 5 were retrospective. The definition of fever ranged from a rectal temperature of 38.0°C or higher measured either at home or at the treating institution to 38.2°C or higher measured at the institution. Patients were evaluated in the emergency department in 8 studies, in either the emergency department or a hospital clinic in 2 studies, and in a hospital outpatient area in 4 studies. Age groupings varied between studies by up to 4 days. Body fluid culture rates varied by study and source between 90% and 100%, with 11 studies reporting that 100% of patients had blood, urine, and CSF cultures performed. One study, included in the analysis of B/BM, reported that all infants younger than 2 months had blood cultures performed and most had CSF cultures performed.⁴⁵

View this table: [in this window] [in a new window] Table 1. Listeria monocytogenes and Enterococcal Infections in Included Studies*

In total, there were 5 infections caused by L monocytogenes and 12 infections caused by enterococci among 5247 infants studied (data regarding urine cultures were missing in 3 studies). No L monocytogenes or enterococcal infections were reported in sources other than blood, urine, and CSF, and there were no L monocytogenes cultures positive in urine specimens. The oldest infant with a L monocytogenes infection was 30 days of age and had both bacteremia and bacterial meningitis. This infant was included in the second month category because the child was in the 29- to 56-day-old infant group reported by Baker et al.³³ Enterococcal infections were found in all 3 body fluids and all age groups.

The prevalence of L monocytogenes and enterococcal infections and the number of children who would need empirical treatment with ampicillin to cover for one of these infections are presented by age in Table 2. The prevalences of B/BM were similar in the first and third months of life and lower in the second month of life (P = .02 for comparison between the 3 months).

View this table: [in this window] [in a new window] Table 2. Prevalence of Listeria monocytogenes and Enterococcal Infection by Age and Source of Infection*

The incidence of perinatal listeriosis is reported to be decreasing.² We did not find evidence of this in our review of febrile infants. When the 3 most recent studies^32-33,41 were evaluated separately, the rates of Listeria infection were similar, with 2 Listeria infections reported in 1048 infants younger than 60 days. Based on 2 of these studies^{32, 41} that evaluated infants younger than 1 month, the NNCs to empirically treat Listeria or enterococcal infections were 313 and 126 for B/BM and SBI, respectively.

COMMENT

Our results, in contrast, support current practice. We believe the use of ampicillin to cover for Listeria and enterococcal infections is still justified in the first month of life. In this age range, 1 in every 439 outpatient infants with fever will have a L monocytogenes infection, which causes high morbidity and mortality, potentially benefiting from expedient therapy with ampicillin. In addition, 1 in every 585 and 345 febrile infants will be appropriately treated for enterococcal B/BM and SBI, respectively. We believe it is reasonable to empirically treat a large number of infants, all of whom we would hospitalize in any event, in an effort to promptly treat those few who have Listeria and enterococcal infections.

Studies included in this review had 1-month age definitions that ranged from 28 to 30 days. Since there was 1 infant 30 days of age with L monocytogenes bacteremia and meningitis, it seems reasonable to choose 30 days of age as a cutoff for empirical ampicillin use.

Between 31 and 60 days of age, we believe the NNCs for B/BM (1544) and SBI (527) may be too large to merit adding ampicillin to empirical therapy for all infants, when empirical therapy is indicated. In this age group, the empirical use of ampicillin in addition to a third-generation cephalosporin would primarily improve coverage of enterococcal infections, most of which would involve the urinary tract. We agree with the recommendation of Sadow et al to add ampicillin if CSF findings are suggestive of meningitis and their alternate proposal to add ampicillin if urine gram stain identifies gram-positive cocci or a gram stain is unavailable but urinalysis is highly suggestive of infection. In addition, we would advocate that ampicillin be considered for any infant who appears very ill or septic, to empirically treat for the rare, but nevertheless possible, Listeria or enterococcal bacteremia.

Surprisingly, the prevalence of reported L monocytogenes and enterococcal infections was greater among infants in their third month of life compared with those in the second month. This may reflect increased rates of enterococcal urinary tract infections in the third month compared with the second month of life. It could also be a result of referral bias, if well-appearing febrile infants in the second month of life are more likely to be referred to the emergency department than are similar infants in the third month of life. Given the small number of studies involving infants in the third month of life, confidence intervals surrounding these estimates were large, although the differences in prevalence rates were statistically significant. Although the numbers needed to treat were similar in the first and third months of life, only enterococcal infections would have been undertreated if ampicillin had not been given empirically in the third month of life. We suggest that empirical ampicillin treatment for infants in the third month of life be determined using the same criteria applied to the infants in the second month of life.

Many authors advocate withholding any antibiotics for low-risk infants after the first month of life. Our study was not intended to determine when empirical treatment is warranted, but rather when ampicillin should be added to an empirical regimen. Institutions where only a subset of febrile infants receive empirical antibiotic treatment will likely have higher rates of bacterial infection in the treated group, likely including an increased proportion of Listeria and enterococcal infections. The NNCs might be lower in this subset.

This review has a number of limitations. The potential benefit of empirical ampicillin use could be overestimated if ampicillin is less than 100% efficacious in treating or ameliorating the course of Listeria and enterococcal infections. The potential benefit could be underestimated if there were Listeria and enterococcal infections that did not result in positive cultures or if other serious bacterial illnesses, such as pneumonia, cellulitis, or osteomyelitis, were caused by these infections. In addition, 5 of the 14 studies had culture rates less than 100%, so they may have underreported the prevalence of bacterial infection. We were unable to evaluate specific characteristics of the infants with positive Listeria or enterococcal cultures, because this information was typically not available.

During the analysis, we were concerned that studies without positive cultures would be more likely to have adequate information for inclusion than would studies with cultures positive for Listeria and enterococcal infections. However, no articles were excluded that identified patients in 1-month blocks and did not specify the ages of the infants with positive cultures. One included study⁴³ could not be analyzed for SBI because it mentioned 2 enterococcal urinary tract infections that could not be categorized by age. On the other hand, 4 studies^{57, 71, 110, 118} were excluded because they did not break down infants into 1-month age groups, and none of these studies had infants with cultures positive for Listeria and enterococcal infections.

Our findings generally support the current practice at our institution, which includes empirical use of ampicillin for febrile infants in the first 4 weeks of life. Empirical ampicillin in this age range is particularly important in institutions such as ours that routinely use this drug in combination with an aminoglycoside, since it will additionally provide improved coverage for group B streptococcus. Infants in the second and third months of life with suspected pyelonephritis also frequently receive treatment with ampicillin in addition to either gentamicin or a third-generation cephalosporin, pending culture results. Our findings in this review may encourage us to consider ampicillin for the occasional febrile infant in the second and third months of life with gram-positive cocci on CSF gram stain or a septic appearance.

What This Study Adds Febrile infants are at risk for serious bacterial infection. Ampicillin is often used in empirical treatment primarily to cover L monocytogenes and enterococcal infections. This systematic review provides evidence to support the common practice of empirical ampicillin treatment for febrile infants younger than 1 month and additionally recommends ampicillin for a selected subset of febrile infants in the second and third months of life.

We thank Robert Davis, MD, MPH, and Fred Rivara, MD, MPH, for their encouragement and thoughtful critique of the manuscript and Gini Scott for her administrative assistance.

Corresponding author and reprints: Julie C. Brown, MD, Children's Hospital and Regional Medical Center, 4800 Sand Point Way NE, PO Box 5371/CH-04, Seattle, WA 98105-0371 (e-mail: jbrow1{at}chmc.org).