 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
a Department of Pediatrics, University of California, San Francisco, San Francisco, California
b Division of Emergency Medicine, A.I. duPont Hospital for Children, Wilmington, Delaware
c Department of Children's Health Services Research, Indiana University School of Medicine, Indianapolis, Indiana
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONREFERENCES |
|---|
39°C without source among well-appearing, 3- to 36-month-old children.
METHODS. A decision analysis was performed to compare the benefits and outcomes of 3 management options (treat: blood culture and antibiotics for all children; test: blood culture and complete blood count for all children, with antibiotics for selected children; observe: no immediate intervention). A hypothetical cohort of 100000 children with fever of 39°C with no obvious source of infection was modeled for each strategy. Using this model, we identified the treatment option that would best suit each parent's preferences, on the basis of parental utilities (from a prior study) for various interventions and outcomes at vaccine efficacies of 0% (ie, no vaccine) and 95%. In addition, we performed survival analyses to assess the morbidity and mortality rates associated with each treatment strategy at various vaccine efficacies.
RESULTS. At a vaccine efficacy of 0%, the majority of parents' preferences suggested the treat option, the strategy with the lowest mortality rate. At a vaccine efficacy of 95%, mortality rates were similar for all 3 management options (
1 in 100000), but parental preferences were still aligned with different options; 50% suggested observe, 42% suggested test, and 8% suggested treat.
CONCLUSIONS. Like physicians, parents have different approaches to risk. With the conjugate pneumococcal vaccine, risks of complications from fever without source are low regardless of treatment strategy. Rather than having a "one size fits all" approach, it is reasonable to incorporate parental preferences into the treatment decision.
Key Words: child • decision-making • fever of unknown origin • parental attitudes • patient-doctor communication
Abbreviations: WBC—white blood cell
Guidelines for the care of febrile children 3 to 36 months of age, without a source of infection,1 have generated much debate.2–5 When Baraff et al1 published guidelines a decade ago, the reported risk of occult bacteremia (3–11%) supported the use of empiric antibiotic therapy for all children with fever of 39°C, with or without a peripheral white blood cell (WBC) count of 15000 cells per mm3. Since the guidelines were issued, however, the Haemophilus influenzae type b conjugate vaccine has virtually eliminated H influenzae as an invasive pathogen among children,6 and it seems that the conjugate pneumococcal vaccine will reduce invasive pneumococcal infections similarly.7,8 In addition, as antibiotic resistance rates increase, physicians are less apt to treat well-appearing febrile children without a proven bacterial infection, to minimize exposure to antibiotics.9,10
The reduction in the incidence of occult bacteremia and the increase in antibiotic resistance rates have generated additional inquiry into the care of young children with fever without apparent source. Assuming high efficacy of the pneumococcal vaccine, recent analyses found that observation alone would be the best treatment option, on the basis of cost and quality-adjusted life expectancy.11,12 However, these analyses do not reflect the complexities that affect physicians' care of febrile children, including each physician's comfort with risk and the desires of the child's family.
No recent studies have formally analyzed the parental perspective on treating febrile children. Although physicians may think that the risk of a serious bacterial infection is sufficiently low to warrant observation alone, parental preferences may not coincide with this management strategy. Although some parents accept many interventions, including blood draws and hospitalizations, to avoid a minute risk of death,13 other parents are more eager to avoid potentially unnecessary interventions.14,15 Therefore, we undertook a study that incorporates parental preferences in evaluating management options for febrile children with no obvious source of infection.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
15000 cells per mm3 receive parenteral antibiotic therapy. This option also carries a small risk of death related to treatment. Because the sensitivity of WBC counts is not 100%, some children with bacteremia go untreated (20–50% of those with bacteremia, depending on the pathogen); because the specificity is <100%, some children receive antibiotics unnecessarily (30% of those without bacterial infection).18–20 (3) Observe assumes that no blood is drawn and no children receive presumptive antibiotic treatment; therefore, any child with occult bacteremia goes untreated initially. No child is at risk of treatment-related death, and no child receives antibiotics unnecessarily. This option carries the risk of death from untreated bacteremia and its sequelae. We assigned parental utilities from a previously published study13 to each possible outcome of the model, ie, blood draw, hospitalization, meningitis with full recovery, meningitis with sequelae, and death. Parental utilities were assessed with the standard gamble utility assessment technique. Parents were presented with outcomes in groups of 3 and were asked to choose between a certain intermediate outcome (eg, hospitalization) and a gamble between some chance of a better outcome (eg, blood draw) and a worse outcome (eg, meningitis with recovery). A question might be as follows: "Would you prefer certain hospitalization, or would you rather take a gamble that your child might have only a blood draw (50% chance) or might have meningitis with full recovery (50% chance)?" After this question, interviewers vary the probabilities in the gamble until the parent is indifferent between hospitalization and the gamble between blood draw and meningitis with recovery. The probability of blood draw (the better outcome) at which the parent is indifferent defines the utility for hospitalization (the intermediate outcome).
Clinical Assumptions
To frame the problem for decision analysis, the following assumptions were made. (1) Each child has no underlying medical problems and exhibits no toxicity (with no signs of meningitis); urinary tract infection and bacterial pneumonia have been ruled out as sources of fever. (2) Each child would have follow-up assessments in 24 hours. (3) Nonbacteremic children would recover without sequelae. (4) A WBC count of 15000 cells per mm3 affects the probability of bacteremia but not the probability of subsequent complications among children with bacteremia. (5) Antibiotics affect the probability that bacteremia would resolve. If bacteremia were to persist, then our model assumed that the pathogen was resistant to the antibiotic used and that antibiotics would not affect the risk of meningitis. (6) We did not model serious bacterial infections other than bacteremia or meningitis. (7) Our model did not include complications resulting from persistent bacteremia or from antibiotic therapy other than death.
Decision Tree
We created a decision tree to model the possible outcomes of fever without a source (Fig 1) by using DATA 4.0 software (TreeAge Software, Williamstown, MA). Branches representing therapeutic options emanate from decision nodes (squares); branches representing chance events (eg, presence or absence of bacteremia) emanate from chance nodes (circles). Each terminal branch (ending in a triangle) represents a possible outcome, ie, (1) death (treatment-related, from meningitis, or from sepsis), (2) survival with sequelae, or (3) survival without sequelae (no bacteremia or full recovery). The proportion of children ending up at each terminal branch was determined by the probability assigned to each chance occurrence leading up to that terminal branch. The baseline probabilities used in this model were derived from published studies and are shown in Table 1.
|
|
This model included Streptococcus pneumoniae, Salmonella species, and Neisseria meningitidis. Although other pathogens are implicated in occult bacteremia (eg, group A Streptococcus, Moraxella catarrhalis, and Escherichia coli), they have been less common than S pneumoniae or Salmonella species and less invasive than meningococcus.16–18,21 Therefore, it is unlikely that including other organisms would change the outcomes of this decision tree significantly.
Probabilities
Baseline Values
The probabilities used at baseline in this analysis are presented in Table 1. For all probabilities, we included the highest and lowest reported values in our sensitivity analysis. Rationale for the choice of baseline values follows.
Prevalence of Bacteremia Among Febrile Children Without Toxicity
The most recent studies reported 1.5% to 2.5% prevalence for pneumococcus (before the pneumococcal vaccine),16–18 and our model assumed 2% at baseline. An earlier study reported a prevalence of 5.7%, but this was among only children who had been assigned randomly22; if nonassigned children were also included, then the prevalence was 3%. Reported prevalence rates for Salmonella range from 0.074% to 0.25%16,18,20,22 and those for N meningitidis range from 0.02% to 0.03%,17,18,20 and our model assumed 0.10% and 0.025%, respectively.
Likelihood of Spontaneous Resolution of Bacteremia
For pneumococcus, we used 90% from the meta-analysis by Rothrock et al23; other studies reported 81% to 94%.21,22 Because Salmonella and N meningitidis are much less common, the values are less definitive and range from 40% to 50% for Salmonella20,24 (we assumed 50%) and from 33% to 57% for N meningitidis25,26 (we assumed 33%).
Antibiotic Efficacy in Clearing Bacteremia
For pneumococcus, we used an efficacy of 95% at baseline, from the large prospective study by Jaffe et al.22 Although other studies using parenteral antibiotic therapy reported efficacy rates of up to 100%,17,21,27 the lower efficacy incorporates increasing antibiotic resistance of pneumococcus. Because Salmonella and N meningitidis bacteremia are less common, the ranges are broad, with reports of 50% to 100% for Salmonella17,20 (we used 50%) and 60% to 100% for N meningitidis, with oral and parenteral antibiotic therapy included17,20 (we used 80%).
Likelihood of Meningitis in Persistent Bacteremia
Because meningitis is a rare outcome, there are few data and very few studies report outcomes with parenteral antibiotic therapy.17,21,27 To facilitate analysis, this model simplified possible outcomes, as follows: a child's bacteremia would resolve or persist (risk affected by antibiotics); if bacteremia persisted, then the child might develop meningitis (risk unaffected by antibiotics). The probability that a child with persistent bacteremia would develop meningitis was derived from reports of children not receiving antibiotics. We assumed that 2.5% of children with pneumococcal bacteremia would develop meningitis, as presented in the meta-analysis by Rothrock et al23 (individual studies reported rates from 0% to 5.1%21,22,28). We found no reports of Salmonella bacteremia progressing to meningitis, with a range of 50% to 100% for N meningitidis25,26 (we assumed 50%).
Likelihood of Meningitis Leading to Death
We assumed 10%, the mortality rate reported by most studies for pneumococcus,28–31 with a range from 6.0%11 to 25.0%.21 Reports on N meningitidis range from 4%20 (used in this model) to 7% among children with clinically apparent infection.32
Likelihood of Meningitis Leading to Severe Sequelae
We used 9.7% for pneumococcus, derived from a large study that reported on moderate and severe sequelae.33 A separate study reporting 25.0% included all sequelae of any severity.31 For N meningitidis, we assumed 3% (reported range: 2–3%32,33).
Likelihood of Persistent Bacteremia (Sepsis) Leading to Death
Because this outcome is so rare, few reports exist and we assumed 0% for all bacteria. The only reported pneumococcal death resulting from bacteremia was in a child with undiagnosed congenital asplenia.16 One study of N meningitidis that reported a 14% sepsis mortality rate involved children with clinically apparent infections,32 not occult disease. Our sensitivity analyses included the range of reported values.
Sensitivity of WBC Count of 15000 Cells per mm3
For pneumococcus, we assumed 80% (studies report a range of 80–86%18,19). For Salmonella, we used 20% (studies report 10–22%17,24,34). For N meningitidis, we used 30% (studies report 22–32%35,36). Although an absolute neutrophil count of 10000 cells per mm3 is more sensitive for predicting N meningitides (sensitivity: 38%), it is less sensitive for pneumococcus (sensitivity: 76%).20 Because pneumococcus is more prevalent, we chose WBC count 15 000 as the better marker.
Specificity of WBC Count of 15000 Cells per mm3
Reports range from 69% to 77%.18,19 We assumed 69% for all bacteria.
Efficacy of Heptavalent Pneumococcal Vaccine
An initial evaluation of heptavalent pneumococcal vaccine reported 97.4% efficacy against pneumococcal serotypes7 contained in the vaccine. The study by Alpern et al37 of children with occult bacteremia found that 97.7% of pneumococcal disease was caused by vaccine serotypes. Therefore, we assumed a maximal vaccine efficacy of 95.2%. Other studies reported that vaccine serotypes cause only 75% to 92% of pneumococcal bacteremia,38–40 and early studies of the pneumococcal vaccine showed it to be only 89% effective when all serotypes were considered.7 However, those studies included overt and occult bacteremia38–40 and children with comorbidities (such as HIV infection and sickle cell disease) that placed them at risk for infection from less-invasive pathogens,38–40 as well as other types of invasive disease such as otitis media.7 Although these studies may not reflect serotype distribution in occult bacteremia, our model likely portrays the best possible scenario for vaccine efficacy.
Likelihood of Treatment-Related Death
Our estimate of 1 treatment-related death per 100000 treated children was based on studies of penicillin-related deaths conducted with young Army recruits.41,42 No large-scale prospective studies exist for cephalosporins, and the rates of treatment-related anaphylaxis and death are not known. Although rare, deaths resulting from cephalosporins have been reported.43
Utilities
In a previously published study, Bennett et al13 conducted computer-based interviews of 94 parents to determine their values for treatment and possible outcomes of occult bacteremia. Parents were first asked to rank, from best to worst, a list of outcomes and interventions related to occult bacteremia, including blood draw, hospitalization, sequelae from meningitis, and death (see Table 2 for full list). Parents then evaluated each outcome/intervention, comparing it with perfect health and the next worst outcome. For example, a parent is asked to compare a blood draw (to test and possibly to treat a child), which ostensibly guarantees perfect health, with the possibility of hospitalization. A parent who wants a 100% chance of perfect health and would accept a definite blood draw to avoid any chance of hospitalization would assign a utility of 1 to blood draw. In contrast, a parent who is willing to accept some risk of hospitalization if perfect health means getting stuck with a needle would have a utility for blood draw somewhere between 0 and 1. After the utility for blood draw is established, the parent would be asked to compare hospitalization (with perfect health) with a risk of meningitis with full recovery (the next worst outcome). Questions continue in this manner to include all outcomes, and results are scaled to adjust to a full range of utilities. Mean and median parental values (known as utilities) are shown in Table 2.
|
We calculated expected utilities for the 3 treatment strategies by using each parent's set of utilities (90 of the 94 original sets of utilities were available for this analysis). The treatment strategy with the highest expected utility represents the strategy most compatible with parental preferences for interventions and outcomes.
Sensitivity Analysis
The decision tree provided the optimal treatment strategy for parents based on their expected utilities. However, because expected utilities depend on assumed probabilities, if assumptions change (eg, if the incidence of treatment-related death is higher or lower than the 1 case per 100000 patients we assumed at baseline), then the strategy most compatible with parental preferences might also change. Sensitivity analysis was the technique used to determine how sensitive outcomes are to changes in assumptions. To perform sensitivity analyses for all 90 parents, we grouped parents according to their optimal treatment strategy (based on expected utilities) at 0% vaccine efficacy and at 95.2% vaccine efficacy. We then calculated mean utilities for each of these groups (Table 3). Using the mean parental utilities for each group, we performed 1-way sensitivity analyses on all other variables, to determine whether changing any probability would change that group's preferred treatment strategy. All sensitivity analyses for clinical variables included the range of probabilities reported in the literature (Table 1).
|
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
39°C without apparent source. At 0% efficacy (all other probabilities at baseline), the smallest number of children died with the treat strategy. Once the vaccine reached an efficacy of 47.1%, testing prevented more deaths than presumptive treatment. When vaccine efficacy was 94.1%, observation alone prevented the most deaths. Although the fewest children died with observation at high vaccine efficacy, this management strategy led to the most cases of meningitis and serious sequelae.
|
Parental Utilities
At a vaccine efficacy of 0%, 84% of parents' preferences favored the treat strategy and 16% the observe strategy. As the efficacy of the vaccine was increased, the treatment strategy most consistent with parental preferences proceeded from treat to test to observe (Fig 3). When vaccine efficacy reached 95.2%, 50% of parents' preferences favored observation, 42% favored testing, and 8% favored presumptive treatment. At 95.2% efficacy, parents in the observe group had the lowest utility for venipuncture, parents in the test group had low utilities for sequelae but high utilities for full recovery from meningitis, and parents in the treat group had low utilities for meningitis with or without sequelae (Fig 4).
|
|
At 95.2% efficacy, observe was a robust result for parents in the observe group and was insensitive to any single probability changing. For parents in the test group, test was a less robust result. This group's mean utilities would favor observe if the risk of treatment-related death were >1 in 70000 or treat if the risk were <1 in 190000. In addition, if the rate of death resulting from meningococcal sepsis were 7% or if the incidence of severe sequelae from meningococcal meningitis were >12.7%, then utilities would suggest treat.
Sensitivity analyses of parental utilities suggested that disutility for blood draw was the utility most strongly associated with treatment preference. For the group of parents whose preferences suggested treat at 95.2% efficacy, if the mean utility for blood draw was <0.99998 (equivalent to a willingness to take more than a 1 in 50000 chance of death to avoid venipuncture), then observe became the strategy with the highest expected utility. Conversely, for parents in the observe group, when there was no parental disutility for blood draw (utility of 0.999995), test became the preferred option. For parents in the test group, slightly lower parental utilities for hospitalization (<0.9892) or recovery (<0.9279) made treat the preferred option, whereas a minute decrease in utility for blood draw made observe the best choice.
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
With this model's baseline assumptions, the risk of death for febrile 3- to 36-month-old children without a pneumococcal vaccine would be 1 in 76000 with the treat strategy and 1 in 18800 with the observe strategy. With a high-efficacy pneumococcal vaccine, the risk of death decreases to 1 in 93000 with treatment and 1 in 172000 with observation. To put these risks in perspective, the annual risk of death in the United States in this age group is 1 in 22000 from a motor vehicle accident and 1 in 27000 from drowning.46
Some parents are willing to accept a >1 in 200 chance of death to avoid venipuncture. Other parents would not accept a risk of death of even 1 in 10000000 to avoid venipuncture. These strikingly different numbers highlight the diverse approaches parents take to risk. Although it may be counterintuitive that parents would risk a 1 in 200 chance of death to avoid venipuncture, other studies of parental utilities, using visual analog scales, found that parents were even more averse to venipuncture.14,15 When a child looks well and the possibility of bacteremia is low but a needle stick is imminent, some parents will understandably opt for observation.
With a successful vaccine, new recommendations will likely be that we observe immunized febrile children.47,48 However, even with a highly efficacious vaccine, a subset of parents will prefer testing or treating presumptively. In addition, perceptions of risk affect physicians' practice styles,49,50 and physicians may be uncomfortable with observation alone, either to avoid the chagrin factor (missing a case of bacteremia) or from fear of a lawsuit.
This model, as with any decision analysis, is limited by the assumptions made in its creation. For example, we did not model antibiotic resistance, because we do not have measures of parental utilities for it and because of the temporal and regional variations in resistance. Including antibiotic resistance could make observe the best choice for most parents at lower vaccine efficacies. However, it is unlikely that including antibiotic resistance or other possible complications would lead to agreement among all parents because parents differ in their approaches to risk. In addition to the assumptions made in constructing the model, the accuracy of the probabilities used within it affect the results of the decision analysis. Although they were well researched and based on existing data, some of the probabilities are difficult to predict, such as the pneumococcal vaccine efficacy, or are controversial, such as the incidence of bacteremia and treatment-related death. The 2% probability of occult pneumococcal bacteremia assumed in this model reflects the low rates seen in 2 large studies.16,18 This prevalence is lower than that in previous studies,17,21,22 likely because Alpern et al16 and Lee and Harper18 included patients who were already receiving antibiotics at the time of blood culture, as well as those immunized recently. We preferred to include these children because recent immunization has not been shown to affect rates of bacteremia51 and the only adverse outcomes (death and meningitis) in the study by Alpern et al16 occurred among children who were receiving oral antibiotic therapy when they presented to the emergency department with fever.
In a recent decision analysis, Yamamoto12 assumed risk of treatment-related death to be between 0.001 and 0.0001, using the "occurrence rate of the negative consequences of treatment perceived by practicing physicians." Our model's baseline incidence of 0.00001 was derived from data for penicillin treatment-related deaths.41 No such large-scale prospective study has been performed for cephalosporins, and the rates of treatment-related anaphylaxis and death are not known. Although rare, deaths resulting from cephalosporins have been reported.43 There is no reason to think, however, that the risk is greater than that with penicillins.52
From a cost standpoint, the study by Lee et al11 suggests that, if the pneumococcal vaccine is highly effective, then observation is the most cost-effective strategy. However, the uncertainties regarding vaccine efficacy, the risks of treatment-related death, and other concerns (such as the possibility of nonvaccine pneumococcal serotypes increasing in prevalence53) highlight the difficulty of promoting a single treatment strategy as categorically best. Most parents accept observation; they do not wish to subject their child to pain to avoid unlikely complications. However, some parents are very risk-averse and prefer to test or to treat presumptively. These parents find venipuncture, in the context of the possible risks associated with bacteremia, much more acceptable than do parents who prefer observation. On the basis of the study by Lee et al,11 with a highly efficacious pneumococcal vaccine, the incremental cost of the test strategy would average approximately $60 per case over the cost of observation alone.
As vaccine efficacy improves, adverse outcomes of fever without source are rare, regardless of the treatment strategy chosen. This model, because it demonstrates that parents vary in their approach to risk, can aid practitioners in understanding parents' resistance to observation alone and can thus inform the physician/parent dialogue regarding treatment options.
|
FOOTNOTES |
|---|
Address correspondence to Kristine A. Madsen, MD, MPH, University of California, San Francisco, Box 0503, 3333 California St, San Francisco, CA 94118. E-mail: madsenk{at}peds.ucsf.edu
This work was performed at the Department of Children's Health Services Research, Indiana University School of Medicine.
The authors have indicated they have no financial relationships relevant to this article to disclose.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  B. W. Trautner, C. G. Macias, and A. C. Caviness High Fever: In Reply. Pediatrics, December 1, 2006; 118(6): 2604a - 2605. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
