Key Questions Addressed in This Report
The University of Ottawa EPC's evidence report on Diagnosis & Management of Febrile Infants (0–3 months) is based on a systematic review of the scientific literature. A technical expert panel helped revise the Key Questions and provide expertise to the review team during the review process.
The Key Questions (KQ) are:
In infants < 3 months old who present with a fever, what are the sensitivity, specificity and predictive values of individual or combinations of clinical features (history including information on the mother's history and previous testing, risk factors, findings on clinical exam, laboratory tests, and formal scoring instruments based on clinical features) for identifying those with serious bacterial illness (SBI)?
How do these findings vary by age within the age range 0 to 3 months?
In infants < 3 months old who present with a fever, what are the sensitivity, specificity and predictive values of individual or combinations of clinical features (history including information on the mother's history and previous testing, risk factors, findings on clinical exam, laboratory tests, and formal scoring instruments based on clinical features) for identifying those with invasive herpes simplex virus infection? How do these findings vary by age within the age range 0 to 3 months?
What is the evidence that clinical features alone, basic laboratory tests (e.g., complete blood count [CBC], urinalysis) alone or the combination are sufficient to identify febrile infants < 3 months who are at low risk of having a serious bacterial illness (i.e., have a high negative predictive value)?
What is the evidence for the potential risks resulting from a delay in the diagnosis and treatment of patients who appear low risk but have a serious bacterial illness?
What is the evidence that clinical features alone, basic laboratory tests (e.g., complete blood count [CBC], urinalysis) alone or the combination are sufficient to identify febrile infants < 3 months who are at high risk of having a serious bacterial illness (i.e., have a high positive predictive value)?
What is the evidence on the benefits and harms of immediate antibiotic (antibacterial and antiviral) therapy and or hospitalization (vs. delaying until diagnostic workup is complete) in patients at high risk of serious bacterial illness?
What is the evidence that the presence of an identified viral infection predicts against a serious bacterial infection?
What is the evidence that the prevalence of serious bacterial illness varies among febrile infants presenting to primary care and emergency practice? What is the evidence that prevalence affects the predictive value of clinical and laboratory findings?
Clinicians base decisions about initial diagnostic work-up and treatment of febrile infants not solely on the infants' medical status but also on their assessments of nonclinical factors (e.g., parental understanding, parents' ability to monitor the patient, access to care). A strategy of initial observation without extensive diagnostic tests or hospitalization depends on confidence that parents will reliably bring the baby back for a timely followup appointment if conditions warrant. How likely are parents whose infants are less than 6 months of age and have fever or other potentially serious medical condition to comply with a provider's recommendation that the parent bring the infant back (to that provider or another) for a return appointment to reassess the condition(s) of concern?
What is the evidence that identifiable parental factors (e.g., education, insurance status, living situation, history of previous visits with the provider, time/distance required to travel to an appointment, etc.) allow a provider to judge the likelihood that a parent will adhere to treatment recommendations such as returning for follow-up if circumstances warrant?
What is the evidence that the clinical setting (community practice vs. emergency department and/or hospital outpatient clinic) in which care is sought independently influences the likelihood of compliance with a return appointment?
Data Sources and Search Strategy
Studies were identified through electronic searches in MEDLINE (1950 to September Week 2 2010, OVID interface), MEDLINE in Process (September 29, 2010, OVID interface), CINAHL (1982 to July Week 2 2008, OVID interface), Embase (1980 to 2010 Week 37, OVID interface), EBM Reviews, Cochrane Central Register of Controlled Trials (2nd Quarter 2010, OVID interface), the Cochrane Database of Systematic Reviews (2nd Quarter 2010, Wiley interface), PsycINFO (1806 to September Week 2 2010) and PubMed (Updated to September 22nd, 2010). Whenever possible, the electronic searches were limited to 1973 onwards, as this was the year that the first study examining bacteremia in a walk-in clinic was published.99
We searched for abstracts in the websites of relevant organizations (i.e., Society of Academic Emergency Medicine, American College of Emergency Physicians, Canadian Association of Emergency Physicians, American Academy of Pediatrics and the Pediatric Academic Societies) to identify any unpublished materials. Additional studies were sought from the authors' personal files and by contacting experts.
The search strategies are presented in Appendix A. The electronic search strategies were developed and executed by two experienced information specialists. The main electronic search strategy (MEDLINE) was also peer reviewed using PRESS (Peer Review of Electronic Search Strategies.100 The searches were combined into a single Reference Manager database and duplicate records were manually deleted, providing a database of unique citations (i.e., titles and abstracts). An update search of the electronic databases was run on October 6th, 2008.
Study Eligibility and Screening
Studies were eligible if they reported the diagnosis of serious bacterial infection (e.g., bacterial meningitis, bacteremia, urinary tract infection) or herpes and/or management of infants (0-3 months of age) with no history of major disease(s), presenting with fever (rectal temperature ≥ 38°C) to a hospital clinic, an emergency department, an acute care clinic, or an outpatient office. Given the lack of relevant evidence found for KQ6 with respect to infants 0–3 months of age, the eligibility criteria were expanded to include children aged 0–6 months. The diagnostic test accuracy results for infants older than 3 months of age reported in some studies were not considered in this review (KQ1a). Such studies were included in the review only if they reported other relevant data (e.g., prevalence of SBI, outcomes related to management of febrile infants).
Reports of studies examining participants from North America, Australia, New Zealand, Western Europe (i.e., Belgium, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Switzerland, and United Kingdom), Northern Europe (i.e., Denmark, Finland, Norway, Sweden), Israel, Hong Kong, Japan, Taiwan, and Singapore were eligible to be included.
The eligibility for inclusion was not restricted by study design (e.g., randomized or nonrandomized controlled trials, case-series, cohort, case-control, or cross-sectional/prevalence studies). Case reports, systematic reviews, cost-effectiveness analyses, articles with no patient data specific to our inclusion criteria (e.g., editorials without any data, decision analyses), and those written in languages other than English were excluded. A list of the citations (i.e., titles and abstracts) that were potentially relevant but written in languages other than English was retained and provided to the technical expert panel for their review and is available upon request. Studies were included regardless of their publication status.
Two reviewers (GN, ACT) independently screened titles and abstracts of all identified bibliographic records by using a study relevance form. Two reviewers (CH, ACT) independently screened full-text reports of potentially relevant records. Discrepancies were resolved by discussion or the involvement of a third reviewer (DM).
Initially, a draft standardized data extraction form was developed by the review team and circulated to the technical expert panel members who provided additional expert input after which the form was accordingly modified. Then, two reviewers (ACT, AT) piloted the modified version of the data extraction form before the actual data extraction process began. Two reviewers (ACT, AT) independently extracted relevant information from the included study reports. Afterwards, a third independent reviewer (FY) verified the extracted data.
Abstracted data included study characteristics (e.g., first author, country of research origin, study design), population examined (e.g., age, ethnicity, mother's demographics), methods used to identify or screen for bacterial or herpetic infection, and treatment or management outcomes of the febrile infants. The diagnostic test results (i.e., sensitivity, specificity, positive and negative predictive values) were directly abstracted when reported. Where possible, the test results were derived from the information provided in studies.
The primary outcome was the set of accuracy indices of an index test (e.g., various laboratory and/or clinical criteria, protocols, laboratory thresholds) against a reference standard (e.g., bacterial culture growth in blood, urine or cerebral spinal fluid, viral culture, molecular testing) in predicting the presence of serious bacterial or herpes simplex virus infections. Secondary outcomes were any events or potential risks associated with a delayed diagnosis/treatment of infants with serious bacterial or herpes simplex virus infection or those associated with immediate treatment (antibacterial or antiviral) for infants classified at a high risk for having a serious bacterial or herpes simplex virus infection. Another outcome of interest was the prevalence of serious bacterial infection in febrile infants stratified by the status of viral infection and a clinical setting of presentation (emergency department vs. outpatient clinic). The proportion of parents' compliance in followup examination visits was also one of the relevant outcomes.
Risk of Bias (Study and Reporting Quality)
The study reports were categorized by study design as follows: randomized controlled trial (including quasi-randomized trials), controlled clinical trial, quasi-experimental (e.g., prepost study), cohort, nested case-control, case-control, cross-sectional, case series, and chart review. Studies reporting diagnostic accuracy data, as well as those for which this data could be derived were classified as diagnostic accuracy studies. Two independent reviewers (ACT, AT) assessed the risk of bias (i.e., study and reporting quality) for the included studies. The reviewers were not blinded to any study details.101
The diagnostic test accuracy studies were assessed using the QUADAS tool.19 This validated instrument consists of 14 items (i.e., questions) asking if a study reported information on the applicability, description of selection criteria, and explanation of study withdrawals. The QUADAS items are rated as “Yes,” “No,” or “Unclear.” For potential convenience and to efficiently summarize the quality data, the reviewers for each study assigned a score of 1 to ‘Yes’ rating and score of 0 to ‘No’ or ‘Unclear’ rating across all 14 items. For example, a study that reported or described information for seven out of 14 QUADAS items received a score of seven. We did not assess the study quality of single arm/single cohort studies (for Q2b and Q3b) and chart reviews, as we could not identify a validated way of conducting such appraisals.
Data Synthesis and Analysis
The identified studies were grouped according to the criteria/protocols (i.e., classification methods, index test) used to predict the risk of serious bacterial infection (or herpes simplex virus) in febrile infants. We did not specify the definition of SBI in this report. Instead, the definitions from original studies were presented. The classification criteria were categorized into the following groups: (1) combined (clinical and laboratory) criteria (Boston, Philadelphia, Rochester, Milwaukee, Yale Observational Score), (2) clinical criteria, and (3) laboratory criteria. Further, the identified formal protocols and criteria were categorized as “Low-Risk” and “Not Low-Risk.”
For each study, two by two tables (i.e., cross-tabulation of infant counts classified by index tests and reference standards used for the diagnosis of serious bacterial or herpes simplex virus infection) were constructed in order to calculate all the necessary diagnostic accuracy parameters (i.e., sensitivity, specificity, negative and positive predictive values) with 95 percent confidence intervals (95 percent CI). Where possible, these parameters were ascertained and calculated for separate types of serious bacterial infection (e.g., bacteremia, meningitis, and bacteremia plus meningitis). The sensitivity of a test may be defined as the probability of an infant being classified in high-risk group given the presence of SBI. Specificity - the probability of an infant being classified in low-risk group given the absence of SBI. Two types of error may occur: one when infants with SBI are classified into low-risk groups (false negatives) and the other when infants without SBI are classified into high-risk groups (false positives). Although it is desirable to have both highly sensitive and specific test, it is not feasible because of the tradeoff between the two indices. Predictive values (negative, positive) of a test are probabilities of having or not having SBI given the risk group an infant was classified into. For example, positive predictive value may be defined as a chance of an infant having SBI given the index test classified this infant into high-risk group. The sensitivity, specificity, negative and positive predictive values for clinical different clinical criteria and/or laboratory thresholds were assembled in tables and were qualitatively compared across studies.
The prevalence with 95 percent CIs of serious bacterial infection in virus-positive and virus-negative febrile infants were ascertained, calculated, and compared by means of prevalence ratios and odds ratios with accompanying 95 percent CIs. The prevalence proportions of serious bacterial infection (or herpes simplex virus infection) were qualitatively compared between the two types of setting (i.e., emergency department vs. primary care) through matching the studies by the country of conduct.
The diagnostic accuracy parameters were pooled using the DerSimonian and Laird random effects model if they were based on the application of the same criteria/protocol in similar populations of infants for predicting the specific type of bacterial infection (e.g., total serious bacterial infection, urinary tract infection, and bacteremia). The degree of heterogeneity across the study results was examined graphically by plotting values of sensitivity and specificity. The assessment of heterogeneity was guided by I2 and Chi-squared statistics and corresponding p-values.20 The potential sources of heterogeneity considered a priori were patient population age (0–28 days vs. > 28 days), prevalence of serious bacterial infection (or herpes simplex virus infection), different index tests (i.e., laboratory/clinical criteria, protocols), and different thresholds for any given index test.
The statistical analyses were performed using Meta-Disc (version 1.4).102
Agency for Healthcare Research and Quality (US), Rockville (MD)
Hui C, Neto G, Tsertsvadze A, et al. Diagnosis and Management of Febrile Infants (0–3 Months). Rockville (MD): Agency for Healthcare Research and Quality (US); 2012 Mar. (Evidence Report/Technology Assessments, No. 205.) Methods.