Antibiotic Use in Febrile Children Presenting to the Emergency Department: A Systematic Review

Introduction: While fever is the main complaint among pediatric emergency services and high antibiotic prescription are observed, only a few studies have been published addressing this subject. Therefore this systematic review aims to summarize antibiotic prescriptions in febrile children at the ED and assess its determinants. Methods: We extracted studies published from 2000 to 2017 on antibiotic use in febrile children at the ED from different databases. Author, year, and country of publishing, study design, inclusion criteria, primary outcome, age, and number of children included in the study was extracted. To compare the risk-of-bias all articles were assessed using the MINORS criteria. For the final quality assessment we additionally used the sample size and the primary outcome. Results: We included 26 studies reporting on antibiotic prescription and 28 intervention studies on the effect on antibiotic prescription. In all 54 studies antibiotic prescriptions in the ED varied from 15 to 90.5%, pending on study populations and diagnosis. Respiratory tract infections were mostly studied. Pediatric emergency physicians prescribed significantly less antibiotics then general emergency physicians. Most frequent reported interventions to reduce antibiotics are delayed antibiotic prescription in acute otitis media, viral testing and guidelines. Conclusion: Evidence on antibiotic prescriptions in children with fever presenting to the ED remains inconclusive. Delayed antibiotic prescription in acute otitis media and guidelines for fever and respiratory infections can effectively reduce antibiotic prescription in the ED. The large heterogeneity of type of studies and included populations limits strict conclusions, such a gap in knowledge on the determining factors that influence antibiotic prescription in febrile children presenting to the ED remains.

In contrast to the above, high antibiotic prescriptions are observed in febrile children (4,5). Guidelines, or new diagnostic approaches have shown to effectively reduce antibiotic prescriptions in primary care (6)(7)(8)(9). This is important because unnecessary antibiotic use increases antibiotic resistance (10,11). In contrast to hospital based studies or primary care settings (11)(12)(13)(14)(15), few studies have been published in emergency department (ED) settings nor do we have valid estimates of potential benefits of antibiotic reducing interventions. Therefore our primary study aim is to assess antibiotic prescriptions for febrile children visiting the emergency department and their determinants. Secondary, we aim to investigate potential interventions that have been proven to be effective in the ED.

Study Characteristics
All descriptive and interventional studies published in 2000-2017 reporting on antibiotic use in children (age under 18) with fever in the emergency department were eligible for this review.

Search Strategy
We searched Embase, Medline (OvidSP), Web-of-science, Scopus, Cinahl, Cochrane, PubMed publisher, and Google scholar for the (analogs of) keywords: fever, antibiotics, emergency department, children and antibiotic prescription. Initially search was performed in 2015 and updated in October 2017 (Supplementary Material 1). References were checked for additional articles to be included.

Inclusion
A screening by title/abstract resulted in potential eligible articles that underwent full text review. Two authors reviewed all articles; any discrepancies were solved by oral agreement between authors.
-Setting: Emergency department; if mixed settings, at least 30% (50 patients minimum) of the population needed to be admitted to the ED.
-Design: observational studies and randomized controlled trials with a minimum of 50 participants. -Outcome: the studies had to report the number or percentage of antibiotics prescribed. -Population: participants under the age of 18; if mixed ages, at least 20% of the population needed to be <18 years (with a minimum of 50) or age specific antibiotic prescriptions had to be presented. Studies on children with specific comorbidities only were excluded. -Fever: at least 30% of all included children needed to have fever or the reason of visit was (reported) fever.

Quality Assessment of Included Articles
To compare the risk-of-bias of all these different study designs all articles were assessed using the MINORS criteria (16). Zero points were given for the item if not reported, one point if reported but insufficient and two points if reported and sufficient. As loss to follow-up was not applicable, due to emergency setting, we have let this particular item out of consideration; the maximum score for studies is 14 or 22 for respectively non-comparative and comparative studies. A maximum score on the MINORS criteria was needed to receive the status of a low risk of bias study (A) (17). For the final quality assessment we additionally used the sample size and the primary outcome. A high quality study was defined by status low risk of bias (A) on the MINORS, antibiotic prescription being the primary outcome and a sample size of at least 500 children. Two reviewers (EV and RO) have independently assessed all included studies. Supplementary Material 2 contains the complete quality assessment.

Data Extraction and Analysis
Extracted data included: Author, year, and country of publishing, study design, inclusion criteria, primary outcome, median (or mean when median not available) age, number of included children. Aiming to invest determinants of antibiotic prescription, we additionally extracted (if available): diagnosis, type of antibiotics, type of physicians, and type of intervention. Due to heterogeneity in participants, outcome measures, interventions and study designs, no statistical pooling but a qualitative analysis was performed (18). Results are presented for the 5 main diagnosis, i.e., fever, AOM, pneumonia, other respiratory tract infections (RTI other) and UTI, with a minimum of 50 cases per diagnostic group required.

Literature Search
We obtained 837 articles by literature search. Screening the full text articles excluded 97 out of 151, which leaves 52 articles for data extraction. Two additional studies were included by reference check of included studies (Figure 1).
Sixteen studies (29%) were considered as high quality and 17 (30%) were considered low quality. In general, observational studies did not describe sufficiently how sample size was approximated. Almost all high quality studies, except one (3), used antibiotic prescriptions as a primary outcome. Table 2 presents the antibiotic prescriptions among the five diagnostic groups we distinguished. Sixteen out of 26 descriptive studies focused on febrile children in general, one paper specifically addressed acute otitis media (AOM) (30), two pneumonia (45,63), four other respiratory infections (RTI other) (19,23,43,57), and one urinary tract infections (UTI) (32). One paper on febrile children also provide separate numbers for pneumonia and UTI (3) and one for AOM (61). Two additional papers focused on respiratory infections and provided separate numbers for pneumonia, AOM and RTI other (44,56).

Fever
Sixteen out of 26 studies focused on febrile children in general, seven of them selected children based on fever without source; five included febrile children based on additional testing ( Table 2). In studies of general febrile populations only, antibiotic prescriptions ranged from 15 to 71% (3,31,35,36,39,42,50,61,71). The lowest prescriptions (15%) came from a study on parenteral empirical antibiotics only (50). Study quality did not influence antibiotic prescription rate.
Three high quality, six moderate quality and two low quality studies reported on SBI rate, which ranged from 7 to 41% (Figure 2) (3, 26, 35-38, 42, 44, 50, 60, 71). As the SBI rate in Khine et al. (42) is similar to antibiotic prescriptions, one may question how SBI is defined. Massin et al. (50) reports on parenteral antibiotics only and may not represent antibiotic prescription in total. Focusing on the remaining eight studies, we observe a trend toward higher antibiotic prescriptions with higher rates of SBI, although not significant.
None of the studies on febrile children in general compared antibiotic prescriptions between countries. In the eleven studies   Fever without source: as defined in corresponding study. ł Antibiotic prescription is given for reported age group, except for Ong et al (57) antibiotic use for all ages is given.

Antibiotic Prescription for Specific Diagnoses
Four studies provided data for antibiotic prescription in AOM, ranging from 88-93%. We could not determine influences of age on prescriptions. Five studies reported on antibiotic prescription in pneumonia, ranging from 69 to 93%. The study with the lowest prescription (3) included children <6 years only compared to the other four (including children in the range of 1-18 years). Antibiotic prescription in RTI other (6 studies) varied on a broader range from 17 to 51%, but could not be related to age. Only two studies provided information on antibiotic prescription in UTI, ranging from 66 to 70%.

Type of Antibiotic Prescription
Nine out of 26 (35%) studies [two high quality (30,56)] reported on antibiotic type (Figure 4). Six studies addressed respiratory tract infections (19,30,43,56,57,63) and five were conducted in the US (19,30,32,57,63). We did not observe a predominance for one antibiotic type for a specific diagnosis or country; amoxicillin was always reported. Studies describing cephalosporin use (n = 7) included both second or third generations.

Interventions for AOM
Interventions with a significant effect on antibiotic reduction were guidelines and the wait-and-see prescription in acute otitis media (AOM). For this latter a significant reduction was found in four articles (three of them with moderate to high quality) (29,34,52,66). High quality study.
Moderate quality study.
Low quality study.

Viral Testing Intervention
Most studies on interventions for reduction of antibiotic prescription addressed rapid viral testing for influenza (RVT, n = 9). Fewer antibiotics were prescribed when the RVT is positive (24,25,27,64), although not confirmed by studies on the impact of RVT use vs. not using RVT in the ED (27,40,58,62). Only one low quality study reported a significant difference for this topic (58). The use of point-of-care testing above testing on indication had only significant benefit in children with proven influenza (33,48). One study reported reduced length of stay, but no effect on antibiotic prescription (48).

Other Interventions
Three high quality studies showed a significant reduction in antibiotic prescription by a guideline for lower respiratory infections or infants with fever (20,21,41). Among two articles on streptococcal A testing, the article with the highest quality didn't find a significant reduction (22,47). Introduction of a clinical pathway for young febrile infants showed reduced time to first antibiotic dose, but did not evaluate the effect on antibiotic prescription itself (53). The use of chest radiographs in particular reduces antibiotics in children with low clinical suspicion of pneumonia (54). For all other interventions no significant reduction was found on antibiotic prescription (46,49,65,69,70).

Interpretation of Main Findings
We observed a highly variable reported antibiotic prescriptions in children presenting to a general or pediatric ED in the five major groups of diagnosis. Studies on a specific diagnosis, such as AOM, pneumonia, or UTI report higher antibiotic prescriptions. However, studies are too heterogeneous to study true effects of determinants. Strong evidence was found for watchful waiting in AOM and implementation of guidelines for fever or respiratory infections to reduce antibiotic use in the ED. Intervention studies report mostly on rapid viral testing for influenzae (RVT) to reduce antibiotic prescription, but its effect is controversial. It is important to note that the high variability in antibiotic prescription observed in our systematic review differ from reported antibiotic prescriptions from literature, or websites (12,73). However, these numbers are based on national or local registries and include in-hospital patients, not reflecting our interest on use of antibiotics in ED settings. Next, not all countries are represented in our systematic review and only Switzerland, USA are represented by more than one study. For the latter two, however we observed high variability in antibiotic prescription within studies of the same country. Even within studies focusing on similar group of diagnoses, we observed a large heterogeneity in their way of patient selection and their type of febrile illness. Therefore, we think these antibiotic prescriptions cannot be considered to be representative for the general population of febrile children in a country.
Limited evidence was found for age effects on antibiotic prescriptions, potentially due to age distribution among study populations. Infants below 2 months are underrepresented in our review. From community studies, we know that pre-school children are more frequently exposed to antibiotic therapy (13).
After exclusion of two outlier studies given their patient selection and outcome definition (42,50), we observed in studies on children with fever a trend toward higher antibiotic prescriptions in studies with higher SBI rates is noticeable. This, however, only explains some variation in antibiotic prescription.
Similar to studies in primary care, watchful waiting intervention seems highly effective for reducing antibiotic use in AOM at the ED (74). Results however are limited to patients above the age of 6 months that did not appear toxic and it is questionable if the study populations were large enough to detect serious adverse outcomes such as meningitis. Although the most frequently studied intervention, rapid viral testing for influenza has no additional effect above testing on indication and controversial evidence was found for its effect. Effects of guidelines are seen in two well-defined groups (respiratory infections or young febrile infants) and including a well-defined implementation plan. Implementation of a clinical decision model to reduce antibiotic prescriptions was only tested in a tertiary pediatric university ED and antibiotic reduction was not a primary outcome of this study (17). All other interventions are not (yet) proven to be effective for reducing the antibiotic prescriptions in children on the ED. Overall the evidence to reduce antibiotic prescription in the emergency department remains limited. We observed a general association between antibiotic prescription and the type of prescriber, i.e., pediatricians prescribe less antibiotics than general physicians may suggest that guideline implementation could be most effective in hospitals with general physicians treating children in the ED.

Limitations
The quality of the studies that reported about fever in general was low to moderate, with only one high quality study (3). Specific drawbacks of study design are included in the MINOR  assessment as a measure of quality. The use of MINORS in combination with the study population and study aim helps to increase the reproducibility of this review and made it possible to compare the different levels of evidence (16).
Most studies did not reported on missing values regarding antibiotic prescription, which could lead to an underestimation of antibiotic prescriptions. In a substantial part of the included papers, antibiotic prescription was not the primary outcome. This may explain some diversity in antibiotic prescriptions, although this was partially corrected for in the quality assessment. This systematic review focuses on prescription of antibiotics in the ED setting. In many European countries, antibiotics are available as over the counter drugs as well (75). This issue is not accounted for by any of the articles, which may lead to a general underestimation of the antibiotic use.
Unfortunately, we observed a large heterogeneity of the studies or had only 1 study per diagnosis group, hampering meta-analysis. Most heterogeneity is caused by specific patient selection (age, setting), by study design (intervention vs. observational cohort study). This also applies to the population of febrile children <36 months that constitute the majority of ED attendances.

Future Research Recommendations
To validly estimate baseline antibiotic prescriptions in children with fever presenting to the emergency department we need observational studies including the general spectrum of febrile children. Being able to determine influences of antibiotic prescription, we should address geographical and cultural influences, differences in setting, adherence area, general patient characteristics, and descriptors of illness severity. Insight in these determinants may help to define targets for intervention to reduce antibiotic prescriptions. Next, this information will contribute to valid power calculations for intervention studies and to generalize effects to other settings.

CONCLUSION
A summary of studies on antibiotic prescription in the 5 main diagnostic groups at the ED did not yield uniform outcomes. There seems to be a trend toward higher antibiotic prescriptions in younger children and for diagnoses that are more often related to bacterial infections. Delayed antibiotic prescription in children with acute otitis media and guidelines for fever/LRTI seem useful to reduce antibiotic prescriptions at the ED. However no strict conclusions can be drawn on the basis of this review because of the large heterogeneity of type of studies and included populations. This means that there is still a gap in knowledge on the determining factors that influence antibiotic prescription in febrile children presenting to the ED. A multicentre study including a wide range of countries on a general population of febrile children would be recommended to provide a valid baseline of antibiotic prescriptions in general, and influencing factors that identify targets for future interventions.

AUTHOR CONTRIBUTIONS
EvdV was responsible for search, dataextraction and writing of the manuscript. HM, SM, and AG contributed to datainterpretation and writing of the manuscript. RO concepted the idea of the paper, supervised search, dataextraction, and writing of the manuscript.