Article Text
Abstract
Objective To explore the risk factors for ward and paediatric assessment unit (PAU) admissions from the emergency department (ED).
Design Prospective observational study.
Setting and patients Febrile children attending a large tertiary care ED during the winter of 2014–2015.
Main outcome measures Ward and PAU admissions, National Institute for Health and Care Excellence (NICE) guidelines classification, reattendance to the ED within 28 days and antibiotic use.
Results A total of 1097 children attending the children's ED with fever were analysed. Risk factors for PAU admission were tachycardia (RR=1.1, 95% CI (1 to 1.1)), ill-appearance (RR=2.2, 95% CI (1.2 to 4.2)), abnormal chest findings (RR=2.1, 95% CI (1.2 to 4.3)), categorised as NICE amber (RR 1.7 95% CI (1.2 to 2.5)). There was a 30% discordance between NICE categorisation at triage and statistical internal validation. Predictors of ward admission were a systemic (RR=6.9, 95% CI (2.4 to 19.8)) or gastrointestinal illness (RR=3.8, 95% (1.4 to 10.4)) and categorised as NICE Red (RR=5.9, 95% CI (2.2 to 15.3)). Only 51 children had probable bacterial pneumonia (4.6%), 52 children had a proven urinary tract infection (4.2%), with just 2 (0.2%) positive blood cultures out of 485 (44%) children who received an antibiotic. 15% of all children reattended by 28 days and were more likely to have been categorised as Amber and had investigations on initial visit.
Conclusions Risk factors for PAU and ward admissions are different in this setting with high reattendance rates and very low proportion of confirmed/probable serious bacterial infections. Future studies need to focus on reducing avoidable admissions and antibiotic treatment.
- Accident & Emergency
- General Paediatrics
- Health services research
- Infectious Diseases
- Outcomes research
Statistics from Altmetric.com
What is already known on this topic?
The rate of serious bacterial infections in children presenting to an emergency department with fever is reported as low (7%).
Acute admissions to paediatric units are paradoxically increasing, but the length of stay is short (<48 hours) highlighting the low-risk nature of the infections encountered.
The National Institute for Health and Care Excellence (NICE) traffic light system for the feverish child are designed to rapidly identify the children aged less than 5 years with an increased risk of serious bacterial infection, but their implementation and compliance at triage has not been evaluated.
What this study adds?
Rates of bacteraemias (0.18%) in a children's emergency department are now substantially lower than reported in NICE 2013 Guidelines.
The risk factors associated with admissions of children to the ward and those admitted to PAU are different.
Admission and reattendances could potentially be reduced by increasing PAU efficiency.
Introduction
Acute febrile illness is a common presentation to the children's emergency department (ED).1 In the largest study on a highly immunised population on febrile children in the ED, the likelihood of serious bacterial infection (SBI) was 7%.2 Invasive infections, such as bacteraemia were present in fewer than 1 in 250 febrile children under the age of 5 years.2 Paradoxically, as rates of childhood SBI continue to fall, hospital admission rates have been increasing, particularly in infants with uncomplicated short-stay admissions for acute infections.3 ,4 Improving the ability of ED clinicians to confidently ‘rule-out’ SBI in the ED could make a valuable contribution to rational decision-making and to reliably predict safe discharge.
The recently updated National Institute for Health and Care Excellence (NICE) guidelines for the management of fever in children under the age of 5 years rely on both subjective and objective clinical evaluations of the child, augmented by biomarkers (C-reactive protein (CRP), white blood cell counts), to define three risk categories; red (high), amber (medium) and green (low).5 These guidelines lack the support of large prospective studies to aid clinical decision-making relating to admission and safe discharge and they are also insufficiently sensitive or specific for the diagnosis of SBI.6 There are alternative models demonstrating additional value of CRP that can discriminate effectively between pneumonia and other SBI but have no impact on patient outcome.7 ,8
What poses an even greater challenge than recognising SBIs is correctly identifying low-risk children. If this is achieved, it would avoid unnecessary hospitalisation and medical interventions. A recent systematic review summarised the published studies using clinical prediction rules with specific emphasis on how they could be used to support the safe discharge of children from the ED.9 To complement the available tools, an evidence-based clinical algorithm integrating point-of-care diagnostic tests, could aid the admission and discharge decision-making process for children with low-risk infections and significantly reduced attendance times.10 ,11 Improved diagnostics may also have a role in rationalising antibiotic prescribing in children and reduction in chest X-rays (CXR) performed.12–14
We present a prospective observational study in a busy tertiary children's ED during one winter season to explore risk predictors for hospital admission in low-risk febrile children including secondary outcomes of antibiotic prescription, investigations performed, time spent in the ED and applicability and accuracy of the NICE guidelines at triage.
Methods
St George's Hospital Children's ED is a tertiary level unit with over 35 000 annual attendances. Prior to commencing the study, ED triage staff were trained in the use of the NICE traffic light system. All children attending the ED were categorised into red, amber or green by the triage nurse at first contact. These categories were then assessed against the objective NICE criteria from the observation charts and the clinical records using internal validation of the models obtained.5 ,15 ,16 All children younger than 16 years of age presenting to the children's ED with fever (≥38°C) or parental report of fever were eligible for inclusion. Only children with significant chronic comorbidities (eg, sickle cell disease, cancer, immunodeficiencies) were excluded. Data were collected from October 2014 to March 2015.
Clinical information was systematically collected using a standard ED pro-forma and included variables from the NICE fever guidelines.5 ,15 Data on time of arrival and transfer discharge were also collected. In addition, antibiotic prescriptions before, during and after ED care were documented. Diagnoses were categorised by clinical syndromes: respiratory, gastrointestinal, urinary and systemic illness (illness affecting more than one system). Bacterial infection was defined as any microbiological cultured confirmed result (urine, blood, cerebrospinal fluid (CSF), respiratory) and/or radiological consolidation in the absence of microbiological respiratory specimens. All clinical care and investigations were performed at the discretion of the attending clinician following local clinical protocols and guidelines.
Outcome measures
The main outcome of the study was to explore the determinants of admission to the paediatric ward and the paediatric assessment unit (PAU) which is adjacent to the main ED, where children can receive care for up to 24 hours. Secondary outcomes included triage agreement with the NICE guidelines,15 ED reattendance rates within 72 hours and within 28 days, readmission to the ward after reattendance, final confirmed diagnosis or exclusion of SBIs (pneumonia, meningitis, urinary tract infection (UTI), bacteraemia) and departmental outcomes such as patient hours in the department (including PAU), antibiotic prescription rates and investigations performed.
Data analysis
Data were entered in ACCESS 2013 (Microsoft) and analysed in STATA (Stata Statistical Software: Release 13. StataCorp LP, College Station, Texas, USA). Crude associations between the main outcomes of interest and potential explanatory candidates were investigated using a univariate analyses. p Values less than 0.05 considered statistically significant. Multivariable logistic regression analyses (multinomial or ordinal) were constructed using backward, forward, stepwise selection or a combination of them. Models choice criteria, such as Akaike information criterion, were used. Criteria for the goodness of fit, such as Hosmer-Lemeshow, were employed to check the appropriateness of the best selected model. Survival analysis was also employed to infer the daily cumulative probability of ED reattendance. Patterns in the missing data have been assessed and multiple imputation techniques17 were performed under missing at random assumption (MAR).18
The most parsimonious model on NICE outcome (red, amber or green) was tested for its generalisability (ie, how accurately it would perform in practice, in a novel situation).16 ,19 In the absence of a completely independent data set, we used a testing–training technique (adapted for an ordinal outcome) by fitting the model on a randomly selected half of the data (training part) and calculating the agreement between the NICE outcome predicted by the model and the documented NICE outcome in the other half (testing data). This process was reiterated 1000 times and the results were summarised as means and IQRs.16
Ethics
Ethics approval was sought from the St George's University Hospitals NHS Foundation Trust R&D department. Informed consent was not necessary because data on routine clinical episodes were collected anonymously as part of a service evaluation of the management of acute febrile illness in the children's ED and there was no intervention.
Results
Data summary
A total of 1117 children with fever were recruited to the study and 1097 children were included in the final analysis (figure 1). The median age of febrile children was 2 years (IQR 1–4 years). Forty-nine per cent of all children were categorised as green (low risk), 45% as amber (medium risk) and 5% as red (high risk) according to the NICE guidelines. Eight hundred and five (73.3%) children attending the ED with fever were sent home. Forty per cent had an investigation performed and 44% had antibiotics either prior to attendance or prescribed in the ED (table 1).
Clinical results and length of stay
Respiratory illnesses were responsible for 73% of the final diagnoses (table 2). Gastrointestinal illnesses (80, 7.2%) were the next most common, followed by systemic illnesses (76, 6.9%) and urinary tract illnesses (38, 3.4%).
Overall rates of culture-confirmed bacterial infections were low (56/1097, 5.1%) and only two (0.18%) were bloodstream infections. Of the microbiological samples taken, 2/140 (1.4%) blood, 52/217 (24%) urine and none (0/18) of the CSF cultures were positive, while 2/18 (11%) throat swab cultures were positive for group A streptococci (GAS) and 22/58 (37.9%) nasopharyngeal aspirates were RSV positive. One patient with a positive blood culture (S. pneumoniae) also had consolidation on CXR.
Children in the main ED had a median length of stay of 3.3 hours (IQR 2.4–4) and those in PAU had a median length of stay of 5 hours (IQR 2.4–9.3).
Compliance with NICE guidelines
Overall, 29% (153/536) of ‘green children’, 46% (217/474) of ‘amber children’ and 73% (36/49) of ‘red children’ had at least one investigation performed in the ED.
Associations with NICE categories by ascending order of severity (red > amber > green) before and after adjusting for potential confounders are shown in table 3. As per internal validation of the regression analysis, NICE categories were in agreement with the model in 70% (767/1097) of children (see online supplementary table S1).
Supplementary table
Results from 1000 simulations of randomly assigned training-testing datasets. The resulting mean of the overall agreement distribution is 69.3% (IQR = (68.1%, 70.5%)).
PAU and paediatric ward admission predictors
Figure 2 summarises the different predictors for PAU and ward admission. Overall children admitted to the ward were more ill appearing and tachypnoeic than those admitted to PAU, who were more tachycardic and likely to have a gastrointestinal illness. Children admitted to both places were more likely to have had an investigation.
Short-stay admissions (<48 hours)
Short-stay ward admissions (<48 hours) were considered as surrogate markers for potentially avoidable admissions. Overall, 62/123 (51%) ward admissions were short stays. When presenting to the ED, children with short-stay admissions were less likely to appear unwell, have investigations performed or receive antibiotics, but more likely to have a respiratory illness compared with those admitted for ≥48 hours (see online supplementary table S2).
Supplementary table
Predictors of short stay in the paediatric ward (< 48 hours).
Reattendance to the ED
Of the children discharged from the ED, reattendance rates at 72 hours and 28 days were 9% (103/1097) and 15% (163/1097), respectively. Of the 103 children reattending at 72 hours NICE categories on initial attendance were 41 (7.6%) green, 56 (11.8%) amber and none red. The highest cumulative probability of reattendance was among children triaged in the amber category who had investigations performed during their initial visit to the ED (figure 3). Children who appeared ill and those with a gastrointestinal presentation were also more likely to reattend. Of the 163 reattenders, 26% (n=42) were admitted to the ward on their second visit to the ED. The rest were discharged home.
Discussion
In this large observational study, we have confirmed the extremely low proportion of culture-confirmed bacterial infections (5%) from our first CABIN study20 and also identified predictors for hospitalisation in previously healthy children with community-onset fever attending a busy paediatric ED. Significant clinical differences were found between children admitted to either the ward or PAU compared with those who were discharged home from the main ED. As per the internal validation of our modelled results (test–training model) for NICE categories, overall, there was a 70% agreement between the NICE category predicted by the test–training model and the documented NICE category at triage. This was mainly due to medium-risk children (amber) being classified as low risk (green), without impact on clinical or departmental outcomes. Given the relatively low rates of confirmed or suspected bacterial infections in febrile children (bacteraemia 0.2%, UTI 4.7%, GAS tonsillitis 0.1% and radiological consolidation 4.6%), antibiotic prescription rates were disproportionally high (44%). Only 26% of children attending the ED with fever were admitted to the PAU (15%) or the ward (11%). Of the 805 children who were discharged home, 9% returned within 72 hours and 15% within 28 days. Children who reattended were more likely to have been categorised as amber risk at triage and had investigations performed at the initial ED visit and, of the reattenders, 26% were subsequently admitted to the ward.
Previous studies have found the NICE traffic light system as having poor discriminatory ability for identifying SBI in febrile children.21 In our cohort, in spite of the low culture confirmed bacterial infections, it was reassuring to note that both bacteraemias found were categorised as high risk or ‘red’ and managed accordingly. We found a 30% discrepancy as predicted by the internal validation (test–training) model and the NICE categories in the ED notes, which is likely to be the consequence of subjective assessment at triage. These differences, however, had no adverse clinical outcomes. Tachypnoeic children who were otherwise well, for example, were often triaged as ‘green’ instead of ‘amber’ or ‘red’. The traffic light system was also easily applied to children of all ages and not restricted to children under 5 years of age. Increasing staff triage training, with the use of NICE at triage, could improve risk classification and might help identify medium-risk children more rapidly (figure 4). Other triage tools, like the Manchester score rely on similar parameters for risk stratification of the patient.22 Overall, our results highlight the usefulness of the NICE traffic light system as an ED triage tool and supports extending its use to nursing and other clinical staff.
The strength of this study relies mostly on the multilevel results from a large cohort of febrile children attending a busy tertiary ED. This gives valuable information on disposition, reattendance and antibiotic use. Our low rates of bacteraemia (1.8/1000 febrile attendances) although with wide CIs because of small numbers (95% CI 0.2 to 6.6/1000) are comparable with the largest recent study on 15 781 febrile young children attending an Australian ED where the risk of bacteraemia was 4/1000 (95% CI 3 to 5/1000) febrile attendances.2 A recent retrospective UK study reported lower community-acquired bacteraemia rates of 0.57/1000 ED attendances in 2011, but their denominator population included all ED attendances rather than febrile attendance.23
We acknowledge as a limitation the lack of information regarding attendances to other hospitals and primary care providers that could increase our estimate of reattendance rates. Also the study ran over 6 months only limiting the seasonality of the associations found. Generalisability to other parts of the UK may not be possible due to the different patient populations included in the different national studies.
There are only limited published retrospective data on PAUs in a UK setting.24 PAUs are a safe alternative to short-stay hospitalisation where stabilisation, observation, diagnosis, treatment and discharge can reasonably be expected within 24 hours,25 such as managing dehydration, monitoring orthopaedic injuries and treating asthma exacerbations.26–30 A recent systematic review describing PAU structure and function in the USA revealed great heterogeneity in mean length of stay (10–35 hours), inpatient admission rates, reattendance and costs.25 The American College of Emergency Physicians introduced the concept of an ‘observation failure’, defined as an admission to the ward after a period of observation or after unscheduled reattendance, with admission rates of >30% from PAU to the ward considered unacceptably high.31 In our cohort, only one child (0.5%) ‘failed’ observation on the PAU and was admitted to the ward, while 24/169 children (14%) reattended the ED within 72 hours of being discharged from PAU. The median length of stay in PAU was 5 hours after an average 4 hours stay in the main ED. Our findings highlight clear differences in the characteristics of children admitted to the ward compared with those admitted to PAU, with the former being sicker overall with systemic illnesses and receiving more clinical interventions, including antibiotics. This separation could help shape future interventions for service improvement, with ample opportunities to optimise the care of children admitted to PAU. We observed, for example, that clinical reviews in PAU were irregular and highly variable in their timings, especially at night. Standardising the timings of nursing and clinical assessments, using predefined checklists, for example, could significantly reduce PAU length of stay and allow for earlier discharges with appropriate safety net advice (figure 4).
Fever-related reattendance rates found in our cohort (15%) were high. Children that reattended had not been categorised as high risk in their initial presentation and were mostly low or medium risk. Unplanned reattendances at 72 hours (9%) were higher than previously described in the UK (4%–5%).32 Exploring potential causes for this, we found associations between baseline investigations and reattendance suggesting that the children were initially sufficiently unwell to warrant a work-up but were subsequently discharged because of normal or pending results. Their reattendance is therefore likely to be a consequence of safety net measures whereby parents are advised to return to the ED if their child's condition deteriorates. However, only 26% of these reattenders were subsequently admitted to the ward. Other possible explanations for reattendance include parental anxiety, diagnostic uncertainty at initial ED visit and low threshold for referrals from primary care among recent ED attenders. Integrating qualitative research to understand the social determinants of readmissions is essential for developing implementation tools (eg, educational interventions) to reduce unnecessary reattendances.
We also identified a clear need to reduce antibiotic prescriptions in the ED. Implementing strategies for antibiotic use reduction include a web-based tool for respiratory infections in primary care.33 ,34 The integration of point-of-care tests in the ED (figure 4) may reduce the diagnostic uncertainty and the amount of investigations, leading to unnecessary use of empiric antibiotics.35
Conclusions
There has been a considerable literature to-date focusing on the now rare SBI in the febrile child. In the resource-rich setting, we suggest that the focus should now shift towards how best to identify the very common low-risk children, who can be safely discharged back to the community without antibiotics. Data from this observational study could be used to develop integrated pathways to reduce PAU and ward admissions in children with self-limiting infections.
Acknowledgments
Acknowledgements We would like to thank the tireless help of Laura Hyrapetian and Sanjana Jaiganesh entering patient data. Also thanks to the nursing paediatric ED staff at St George's Children's ED for their continuous recruitment of children for the CABIN study.
References
Footnotes
Twitter Follow Adam Irwin at @adamdirwin
Contributors MS, JT and RB conceived the work; AB, RS collected and entered the data; ICS performed the statistical planning and main analyses AB, ICS, AI, analysed the data; AB wrote the manuscript; SL, MS, JT and RB reviewed and edited the manuscript.
Funding This work was supported by the Wandsworth Clinical Commisioning group, grant number 13-14-060.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.