Management of the young febrile child: a commentary on recent practice guidelines.

Abstract

Recently published “practice guidelines”12 and randomized antibiotic trials34 reflect a climate of increased diagnostic testing, more frequent treatment, and more invasive (ie, parenteral rather than oral) treatment of febrile children 3 to 36 months of age. For children in this age group with a temperature ≥39.0°C, the guidelines12 suggest a white blood cell (WBC) count and provide two options with respect to obtaining a blood culture: all such children or those whose WBC count is ≥15 000/mm2. Culture of urine obtained by catheterization or suprapubic aspiration is recommended for all boys <6 months and all girls <24 months. The guidelines recommend empiric treatment with ceftriaxone, once again with two options: treat all such children or those whose WBC count is ≥15 000/mm2. These practice guidelines are based on a meta-analysis that pooled data from both randomized controlled trials and observational (nonexperimental) studies of clinical outcomes in young febrile children, and on the views of an expert panel chosen by the senior author.12 Although the guidelines have not been officially endorsed by any professional organization, they were developed by authors who are widely recognized in the field and thus could have an important impact on both clinical practice and health care policy.The clinical setting is that of a child with acute onset (≤4 days) of fever who does not appear “toxic” (ie, seriously ill) and has no apparent focus of bacterial infection (otitis media, pneumonia, osteomyelitis/septic arthritis, lymphadenitis, cellulitis, dysentery-like enteritis, or meningitis) after a history is obtained and a physical examination is performed. When confronted with such a child, the clinician must make a series of decisions:It is hardly surprising that, faced with these multiple complex decisions, the potential threat of litigation, and different policies for follow-up, physicians vary widely in their practice, particularly between those who work in private office settings vs those who work in hospital emergency departments.56 Although some practitioners may welcome practice guidelines that help them navigate the complexity of the clinical issues and reduce their uncertainty7 and their perception of liability, one recent study8 based on simulated case scenarios found little compliance with the recently published guidelines.12The diagnostic tests called for in the guidelines are difficult to obtain for many physicians who practice in private office settings. More important, we believe that both the testing and the treatment may do more harm than good to the young febrile children and their families who are subjected to it. In the remainder of this commentary, we critically review the evidence upon which the recent guidelines are based and propose instead an approach that involves far less testing and no empiric antibiotic treatment, ie, one much closer to that used by many office-based practitioners. We shall argue that such an approach is more than defensible; we maintain that it is actually preferable, not only for the private office but for the emergency department as well.The diagnostic tests recommended in the guidelines12 are the complete blood count (CBC), blood culture, and urine culture. Several of the recommendations are keyed to the results of the CBC, with recommendations to perform a blood culture and/or to begin empiric antibiotic treatment based on the presence of a WBC count ≥15 000/mm3. It has been repeatedly shown that high WBC counts are significantly more common (approximately two- or threefold) in children with bacterial infections than in those with viral infections.9-16But the value of a diagnostic test depends not on whether the association between the test result and the condition of interest (eg, bacteremia) is statistically significant in a group of children, but rather on the predictive value of the result in an individual child. In particular, if the test is positive (or negative), how likely is it that the child has an occult bacterial infection? Because viral infections are far more prevalent than bacterial infections, the positive predictive value of an elevated WBC count for an occult bacterial infection is rather low (8% to 15% in most studies). Tversky and Kahneman17 have documented the widespread failure to adequately consider prior probability (ie, prevalence) in weighing the evidence for and against an uncertain proposition. This common error in probability assessment prevents many clinicians from appreciating the fact that the vast majority of young febrile children with a high WBC count do not have an underlying bacterial infection as a cause of their fever.16 Even among the true positives, ie, those with a high WBC count who do have an occult bacterial infection, most will have an infection that is likely either to clear spontaneously (bacteremia) or to respond to treatment without serious sequelae even if the diagnosis is delayed (pneumonia, cellulitis, UTI). The positive predictive value for more serious infections (meningitis, osteomyelitis, septic arthritis) is at least several orders of magnitude lower.The value of the blood culture itself is questionable. The primary concern is not bacteremia per se, but rather those few bacteremic children who subsequently develop serious focal infections. Many children who develop meningitis or other serious sequelae of their bacteremia will already have done so by the time the blood culture is recognized as positive.18 Moreover, serious sequelae of occult bacteremia are increasingly uncommon. In the past, many of the bacteremic children who developed meningitis or other sequelae were those with Haemophilus influenzae type b (Hib) bacteremia,19-24 which has become rare in the era of routine immunization with Hib conjugate vaccines.25-27Unfortunately, a management strategy that includes obtaining a blood culture as part of the initial assessment can lead to unnecessary hospitalization. Observational studies and randomized trials of young febrile children indicate that the majority of initially bacteremic children who remain febrile do not have persistent bacteremia, even if they did not receive antibiotics at the initial visit.342829 Thus, if the physician decides to hospitalize and treat all initially bacteremic children who remain febrile at follow-up, most such children will be hospitalized and treated unnecessarily.2930 It is precisely the knowledge that such children were bacteremic at the initial assessment that results in excessive hospitalizations. Without such knowledge, few clinicians would hospitalize children who remain febrile but do not appear seriously ill. Hospitalization is considered far more undesirable by young children's parents than by physicians.31Apart from the CBC and blood culture, the recent guidelines recommend a culture of urine obtained by catheterization or suprapubic aspiration for all boys <6 months and all girls <24 months. Catheterization and suprapubic aspiration cause discomfort, however, and the culture results are not usually available for 24 to 48 hours. Although routine urinalysis may not be sufficiently sensitive or specific to make the diagnosis of urinary tract infection reliably,32-34 two recent studies (from the same group of investigators) suggest that “enhanced” urinalysis based on a trained technician's use of a hemocytometer and Gram stain to examine an unspun urine sample may improve both the sensitivity and specificity for diagnosing UTI in this age group.3536 Preliminary evidence suggests that children with positive urine cultures but without pyuria appear to be at little risk of pyelonephritis and may represent cases of asymptomatic bacteriuria rather than of true urinary tract infection.3537 If these data are confirmed in future studies based on routine urinalysis (as performed in most office and hospital laboratories), a negative urinalysis on a clean-voided (bag) specimen may prove sufficient to obviate the need for urine culture obtained by a more invasive technique. Moreover, although a urine culture of a bag specimen is associated with a high risk of either contamination or inconclusive result,38-43 a negative result is strong evidence against a UTI.In arguing against the routine performance of the CBC, blood culture, and bladder catheterization or suprapubic aspiration, we do not dispute that liberal blood and urine testing have occasionally led to earlier relief of symptoms and to prevention of complications, perhaps even of death. But at what price? Apart from the monetary cost of the diagnostic tests themselves, what are the risks to both the child and his or her family? These risks include the pain and discomfort of the diagnostic procedure, the waiting time before the procedure is performed and until the results are received, the need for repeat cultures of blood and urine [due to transient (ie, already resolved) bacteremia or contaminated initial blood or urine specimens], and unnecessary hospitalization and treatment with intravenously administered antibiotics (due to false-positive urinalysis or to contaminated blood culture).7 These risks and inconveniences are hardly life-threatening and are unlikely to be associated with long-term morbidity. At the individual level, they pale before the potential benefits of the diagnostic tests. After all, what are the pain and waiting time associated with a venipuncture compared with the prevention of either death or serious morbidity?But risks and benefits cannot be compared directly against one another, because they do not occur with anything approaching similar frequencies. Every child who receives one or more of these tests will experience pain and discomfort, and every family will have to wait for the test to be performed and for the physician to receive the results. By contrast, prevention of either death or serious morbidity is extremely rare.Moreover, recent studies suggest that physicians and parents have considerably different values for both the risks and the benefits of testing.3144 Parents emphasize the short-term risks of tests (particularly the associated pain or discomfort) and the possibility of diagnostic error, whereas physicians give considerably greater weight to rarely-occurring serious morbidity and long-term adverse sequelae of infections. Thus, even if physicians were capable of adequately considering each of the risks and benefits of diagnostic testing, their decisions might well not be optimal for children and their parents because of the fundamentally different values they place on those risks and benefits. Physicians may also obtain diagnostic tests because they believe that testing may protect them against possible litigation, although we are aware of no evidence to support this belief. The same tests are also commonly ordered in Canada (where malpractice claims are far lower in frequency and magnitude than in the United States), and their use is influenced by practice background and experience and the presence or absence of a clinical trainee.45Can serious complications of bacteremia be prevented in outpatients by treating young febrile children with antibiotics (“expectant” therapy)? Investigators have attempted to answer this question using two approaches: observational studies that retrospectively compare outcomes of treated vs untreated children with occult bacteremia,18-24 and experimental studies (randomized controlled trials) of expectant treatment.342846 Both approaches have been fraught with difficulties in design, analysis, and interpretation.The observational studies report that bacteremic children who were initially treated with antibiotics developed fewer “new” foci of infection than did children who did not receive antibiotics. Observational study designs are inherently biased toward finding a positive effect of treatment, however, because at baseline the children in the two groups are not equally likely to develop the subsequent outcomes (new foci of infection). Children in these studies were not assigned randomly either to receive or not to receive antibiotic treatment. In particular, those who were treated often already had the outcomes of interest (eg, pneumonia or otitis media–which is why they were treated27), so the probability that theysubsequently would develop new foci of infection was substantially lower than that of the untreated children.Because most of the published studies in this area have come from tertiary-care emergency departments or walk-in clinics, almost all treated children had some identified focus of infection at the initial visit. Some of the untreated children, however, may well have had an unrecognized focal infection, such as pneumonia (a chest radiograph may not have been performed) or otitis media (the tympanic membranes may have been difficult to visualize or were thought to be erythematous from crying). It is therefore not surprising that children with undetected focal infections who were not treated with an antibiotic were more likely to remain symptomatic and to have the focal infection detected at follow-up than children who received an antibiotic. Moreover, some untreated patients with more serious focal infections (eg, epiglottitis, septic arthritis) who were not diagnosed at the initial visit were classified as having developed “new” foci of infection at follow-up.19 Finally, several observational studies grouped meningitis with much less serious outcomes such as otitis media and pneumonia; in fact, treated and untreated children showed little difference in risk of developing meningitis.Randomized clinical trials are more likely to yield a scientifically valid answer to the question of whether expectant antibiotic treatment is effective. Nonetheless, all of the published trials (see Table) have had methodologic problems.47 In particular, all four trials limited their statistical analyses to children who later proved to have had bacteremia at the time they were enrolled. Such analyses are simply incorrect; they violate the epidemiologic maxim to “analyze what you randomize.” The proper denominator for such comparisons is the number of febrile children randomized. The treatment of children known to be bacteremic is not controversial; all such children should receive antibiotics parenterally. The pertinent clinical question is whether the benefits of treating all young febrile children (most of whom have self-limited viral infections) outweigh the financial and human costs and the adverse side effects of such a strategy. As we shall see, the trial investigators were sometimes able to obtain statistically significant differences between the groups only by analyzing the subgroup with occult bacteremia (<3% of the study sample), despite the smaller sample size used in their statistical tests. Unlike sociodemographic or clinical characteristics that are identifiable a priori (eg, age, sex, temperature, or presenting signs and symptoms) and that could, therefore, serve as a basis for legitimate subgroup analysis, the presence or absence of bacteremia cannot be ascertained at the time the physician must decide whether to treat or not to treat. Moreover, an analysis restricted to bacteremic children ignores the outcomes in the remaining 97% of children who were randomized and treated.Methodologic issues aside, a clinical example will demonstrate why the analyses should not be limited to the subgroup with bacteremia. A febrile child is evaluated in an emergency department, has blood taken for culture, is treated expectantly with an antibiotic, and is sent home. Although the initial blood culture is negative, the child subsequently develops bacterial meningitis. (This is a plausible scenario, because blood cultures are not 100% sensitive48and bacteremia does not necessarily develop early in a febrile illness and may be intermittent; indeed, most experienced clinicians have seen such patients.) If such a child had been enrolled in these clinical trials, she would not have been included in the statistical analyses, even if she developed meningitis.Carroll and colleagues46 reported the results of a small (n = 96) randomized trial of expectant antimicrobial therapy (benzathine penicillin administered intramuscularly, followed by penicillin V administered orally) vs no treatment in 96 children 6 to 24 months of age with a temperature ≥40°C and elevation of either the WBC count or the erythrocyte sedimentation rate. Although the investigators concluded that the treatment was effective in preventing complications of bacteremia, the data analysis was based only on the 5 bacteremic children in each group; 4 of 5 children who received penicillin were “improved” (afebrile with negative follow-up blood-culture) 24 to 48 hours later vs 0 of 5 untreated children. Such an analysis yields a 2-sided P value of .048 by Fisher exact test, but the correct analysis based on all 96 randomized subjects yields a P value of .101. Moreover, because no placebo control was used, neither the investigators nor the parents of the subjects were blind to treatment allocation. Two of the untreated bacteremic children had pneumococcal meningitis, but nonblinding could have influenced the two diagnoses in the remaining 3 of the 5 who were “unimproved” at follow-up: otitis media or persistent fever. Moreover, if the analysis is based on the more clinically relevant outcome of all new focal bacterial infections regardless of presence or absence of bacteremia at the initial visit, the proportions of 7/50 (treated) vs 10/46 (untreated) yield a relative risk [RR] (and 95% confidence interval) of 0.64 (0.27–1.55) and an uncorrected χ2 value of 0.98 (P = .321).Jaffe et al28 randomized 955 children 3 to 36 months of age with temperatures ≥39°C to receive amoxicillin or placebo in a double-blind manner. No statistically significant differences in outcome were observed between the two groups, although the rarity of focal complications of bacteremia (2 of 507 patients in the amoxicillin group and 1 of 448 in the placebo group) provided insufficient statistical power to rule out a benefit of amoxicillin. Although bacteremic children who received amoxicillin experienced significantly larger mean decreases in body temperature at the time of follow-up, no temperature data were presented for the nonbacteremic children. No child in either group developed meningitis.Bass et al3 randomized 519 children 3 to 36 months of age with temperatures ≥40°C (or ≥39°C plus a WBC count ≥15 000/mm3) to receive either ceftriaxone intramuscularly or amoxicillin/potassium clavulinate orally.3 Of the 509 children whose temperatures at 24 hours were recorded, those treated with ceftriaxone were significantly less likely to be febrile than those treated with amoxicillin/potassium clavulinate (49 of 245 vs 81 of 264) [RR = 0.65 (0.48–0.89)]. The single case of meningitis (due to Hib) occurred in a child who had received ceftriaxone, although the organism was cultured from cerebrospinal fluid (CSF) obtained before treatment.In the largest and most recent trial, which has been widely cited to support the use of expectant treatment of febrile children with ceftriaxone, 6733 children 3 to 36 months of age with temperatures ≥39°C and no apparent bacterial focus of infection (or with otitis media) were randomized to receive either a single dose of ceftriaxone (50 mg/kg) intramuscularly or a short course of amoxicillin orally.4 Unlike the three previous trials, no outcomes were reported according to treatment among the majority of children without bacteremia. To illustrate the importance of reporting outcomes by the proper denominator, the difference in persistent fever among bacteremic children was statistically significant [RR = 0.75 (0.58–0.97)] when analyzed (as the authors did) as a proportion of bacteremic children, but not [RR = 0.81 (0.55–1.19)] when analyzed as a proportion of all treated children. Moreover, 370 (5.7%) of the nonbacteremic children (both treatment groups combined) developed new focal bacterial infections, which is similar to the 6/91 (6.6%) overall incidence reported among the bacteremic children who received amoxicillin, although none of the nonbacteremic children developed culture-positive meningitis or other serious bacterial infections.Fleisher et al4 reported their results in terms of both “definite” and “probable” bacterial infections, based on a priori clinical and laboratory criteria. These criteria, however, produced an intrinsic bias in favor of ceftriaxone; a positive blood or CSF culture at follow-up was required for “definite” meningitis, and a positive culture or Gram stain from a bone specimen was required for “definite” osteomyelitis. Because ceftriaxone concentrations in the CSF, bone, and other sites are far more likely to exceed the minimum inhibitory concentration for potential pathogens and to do so for much longer than amoxicillin concentrations, treatment with ceftriaxone is more likely to result in negative cultures in a child who develops a focal infection such as meningitis. Children who received amoxicillin and developed a bacterial infection were therefore more likely to be classified in the “definite” category. But would any sensible clinician consider the 2 children in the ceftriaxone group with Hib bacteremia (1 with 111 white cells per mm3 and a positive latex agglutination test for Hib in the CSF, the other with a CSF white cell count of 1300/mm3) not to have “definite” meningitis? If only “definite” bacterial infections were considered, the children who received ceftriaxone had a lower incidence (0 of 3333) than those treated with amoxicillin (5 of 3347), a difference that borders on statistical significance. The difference in favor of ceftriaxone was far from statistically significant, however, when probable bacterial infections (which included the two above-mentioned children with clinically apparent, but culture-negative, meningitis after treatment) and definite bacterial infections were combined. Moreover, one of the “definite” cases of meningitis in the amoxicillin group occurred in a child whose CSF obtained at the initial visit was already (ie, before treatment) positive for pneumococcus.Of the 5 “definite” or “probable” cases of meningitis that occurred in this large trial, 4 were due to Hib, an infection that has been virtually eliminated since the introduction of conjugate vaccines.25-27 Consequently, even if one accepted the conclusions of the investigators, the data are now outdated, because the epidemiology of occult bacteremia has changed dramatically since the study was completed. Indeed, the risk of meningitis (the major complication of occult bacteremia that is of concern, because most of the other “complications,” such as cellulitis or pneumonia, can be easily treated when they become apparent and are rarely associated with serious sequelae) among children with occult bacteremia has declined dramatically, because the majority of cases of bacterial meningitis that developed in such children had been due to Hib.21Unfortunately, the recently published guidelines are based on a meta-analysis of all published data on the efficacy of empiric antibiotic treatment, ie, an uncritical pooling of data from observational studies and randomized trials. Such an approach ignores the inherent bias in observational studies (discussed above) toward finding an apparent benefit of treatment. In our view, the evidence from the four randomized trials does not demonstrate a benefit of either orally or parenterally administered antibiotics in reducing the risk of meningitis or of other serious bacterial infections. Antibiotic treatment (particularly intramuscular ceftriaxone) does appear to lead to more rapid defervescence. But we question whether a shorter duration of fever is sufficient to justify expectant treatment, particularly since unrecognized focal infection (eg, otitis media or pneumonia) present at the initial visit might well explain how this apparent benefit arises. Finally, any alleged benefit of antibiotic treatment may not be relevant for the post-Hib conjugate vaccine era.What are the risks of routine expectant treatment of young febrile children with antibiotics? In addition to the substantial financial and human costs, antibiotics have predictable as well as idiopathic adverse side effects.49 In the trial by Jaffe et al,28 diarrhea tended to occur more frequently in the amoxicillin group than in the placebo group. In the large trial by Fleisher et al,4 ceftriaxone recipients were more likely than amoxicillin recipients to develop a rash. Although none of the cases of diarrhea or rash in these studies appears to have been severe, two recent reports describe fatal hemolytic anemia in children (both with serious underlying conditions) who had received a single intravenous dose of ceftriaxone.5051 In addition, because it is more difficult to interpret the results of CSF analyses of children treated with an antibiotic, a modest pleocytosis at the time of follow-up is more likely to result in hospitalization and parenteral treatment of a child for suspected meningitis when it is not necessary. Moreover, use of antibiotics selects for resistant organisms5253 and is an important risk factor for infection by such organisms, including salmonellae54and penicillin-resistant pneumococci.5556 Perhaps most importantly, routine expectant antibiotic treatment could jeopardize thoughtful assessment, individualized management, and close follow-up of the febrile child,4749 accentuate “fever phobia” among parents5758 and physicians,59 and encourage the all-too-common practice of “treating fever” with antibiotics.We return to an issue touched on briefly at the beginning of our commentary: the large gap in management strategies between practitioners who work in emergency departments and those in office practice.56 Our own experience indicates similar differences in Canada, despite universal health insurance and the frequent use of hospital emergency departments as convenient “walk-in” clinics by families from all socioeconomic strata. Because there is no evidence that young children attending private offices and emergency departments differ with respect to the prevalence of bacteremia; the risk of serious sequelae of bacteremia; the sensitivity, specificity, and positive and negative predictive values of the CBC and other diagnostic tests; or the efficacy of antimicrobial therapy, it is difficult to defend such discrepant management strategies. Instead, the discrepancies suggest that inappropriate decisions are being made for some children.Aggressive management in the emergency department is often defended by citing the physician's usual lack of previous contact with the child and family and her consequent uncertainty about the adequacy of parental surveillance and compliance with follow-up. Serious doubts about the parents' ability to observe and respond to their ill child can and should affect the physician's management. But many emergency departments have developed efficient systems of telephone follow-up by nursing and/or physician staff that work well for the vast majority of parents and children, with repeat visits arranged as needed. Diagnostic testing and “blind” treatment are poor substitutes for, and could even discourage, scrupulous follow-up. Repeated observation over time at the initial visit, and return visits if the fever persists or other symptoms or signs develop, are likely to be more helpful.Finally, if obtaining a blood culture and treatment with intramuscular ceftriaxone are preferred for “nontoxic” children without a focus of infection, why are they not also preferred for similarly nontoxic children with identifiable bacterial foci of infection such as otitis media or pneumonia? The prevalence of bacteremia is at least as high in children with otitis media or pneumonia as in those without a bacterial focus of infection.222360-62 On the basis of what evidence can one defend this “double standard” of aggressive testing and parenteral antibiotic treatment of unknown infections vs no testing and oral treatment of known focal infections?We recommend that young febrile children be carefully assessed for bacterial foci of infection. A child who appears toxic should receive appropriate cultures and diagnostic tests, hospitalization, and intravenous antibiotic treatment. If no focus is found and the child appears well, diagnostic tests other than urinalysis are not routinely indicated, and no antibiotic treatment should be given. Close follow-up should be ensured, which will permit subsequent clinical and laboratory evaluation, hospitalization, and/or antibiotic treatment as indicated by worsening or persistence of symptoms and signs of infection. If the physician is concerned about the adequacy of parental follow-up, repeated observation over time in the office or emergency department is often helpful, and hospital admission may occasionally be necessary.These recommendations are based on our reading of the currently available evidence; future research may yield findings that mandate a change in diagnostic or therapeutic management. In particular, if conjugate pneumococcal vaccines currently under development prove effective, adding their use to that of conjugate Hib vaccines should substantially reduce the young febrile child's risk of bacteremia and hence obviate the need for additional studies (and commentaries like this one) on its diagnosis and treatment.We thank Drs C. Davenport Cook, Michael Gerber, John Leventhal, Paul McCarthy, and Ellen Wald for their helpful comments and suggestions on previous versions of this manuscript.