Article Text

Download PDFPDF

Validation of two algorithms for managing children with a non-blanching rash
  1. F Andrew I Riordan1,
  2. Laura Jones2,
  3. Julia Clark3
  4. on behalf of the Non-Blanching Rash Audit Group
    1. 1Department of Child Health, Birmingham Heartlands Hospital, Birmingham, Alder Hey Children's Foundation NHS Trust, Liverpool, UK
    2. 2School of Clinical Medical Sciences (Child Health), University of Newcastle, Royal Hospital for Sick Children, Edinburgh, UK
    3. 3Great North Children's Hospital, Newcastle, Lady Cilento Children's Hospital, Brisbane, UK
    1. Correspondence to Dr A Riordan, Department of Paediatric Infectious Diseases, Alder Hey Children's Foundation NHS Trust, Eaton Road, Liverpool L12 2AP, UK; Andrew.riordan{at}alderhey.nhs.uk

    Abstract

    Background Paediatricians are concerned that children who present with a non-blanching rash (NBR) may have meningococcal disease (MCD). Two algorithms have been devised to help identify which children with an NBR have MCD.

    Aim To evaluate the NBR algorithms’ ability to identify children with MCD.

    Methods The Newcastle-Birmingham-Liverpool (NBL) algorithm was applied retrospectively to three cohorts of children who had presented with NBRs. This algorithm was also piloted in four hospitals, and then used prospectively for 12 months in one hospital. The National Institute for Health and Care Excellence (NICE) algorithm was validated retrospectively using data from all cohorts.

    Results The cohorts included 625 children, 145 (23%) of whom had confirmed or probable MCD. Paediatricians empirically treated 324 (52%) children with antibiotics. The NBL algorithm identified all children with MCD and suggested treatment for a further 86 children (sensitivity 100%, specificity 82%). One child with MCD did not receive immediate antibiotic treatment, despite this being suggested by the algorithm. The NICE algorithm suggested 382 children (61%) who should be treated with antibiotics. This included 141 of the 145 children with MCD (sensitivity 97%, specificity 50%).

    Conclusions These algorithms may help paediatricians identify children with MCD who present with NBRs. The NBL algorithm may be more specific than the NICE algorithm as it includes fewer features suggesting MCD. The only significant delay in treatment of MCD occurred when the algorithms were not followed.

    • Infectious Diseases
    • General Paediatrics

    Statistics from Altmetric.com

    What is already known on this topic

    • Meningococcal disease often presents with a non-blanching rash (NBR).

    • Only 10% of children presenting with NBRs have meningococcal disease.

    • Paediatricians need guidance to know which children with NBRs should be treated with antibiotics.

    What this study adds

    • Paediatricians do not treat all children with NBRs with antibiotics.

    • Two algorithms identified children with meningococcal disease (MCD) presenting with NBRs.

    • The Newcastle-Birmingham-Liverpool (NBL) algorithm may be more specific than the National Institute for Health and Care Excellence (NICE) algorithm as it includes fewer features suggesting MCD.

    Introduction

    Meningococcal disease (MCD) is a significant cause of morbidity and mortality in children.1 Early identification and treatment of children with MCD may improve outcome.2 ,3 One of the characteristic signs of MCD is a non-blanching rash (NBR), although this may be absent in up to 20% of cases.4 Paediatricians are thus concerned that children who present with an NBR may have MCD. There is an increased public and professional awareness of the importance of NBRs making this a common presentation to paediatricians.

    However 90% of children presenting to hospital with an NBR do not have MCD5 and petechiae can be found in 3% of well infants.6 The management of children presenting with NBR thus remains controversial.7

    Missing a case of MCD could have devastating consequences, so some have suggested that all children with NBR should be treated with parenteral antibiotics for MCD.4 Others argue that this is unnecessary8 and that prudent use of broad-spectrum antibiotics is to be recommended, especially as this may decrease antibiotic resistance.9

    Criteria were previously dveloped to help identify children with an NBR who had MCD.8 However these criteria may not identify all children with MCD presenting with an NBR.10

    We had previously used the available literature to develop an algorithm to manage children with NBR presenting to paediatricians in hospital.10 We hoped this would help clinicians make safe clinical decisions when identifying those children with NBR who do not require treatment. Subsequently National Institute for Health and Care Excellence (NICE) guidance on bacterial meningitis and meningococcal septicaemia has been published.11 This includes a slightly different algorithm for managing children presenting with NBR (using fever as an extraclinical feature and limiting blood tests to those with fever).

    The aim of this study was to validate both algorithms on a large cohort of children presenting to paediatricians with NBR.

    Methods

    The first algorithm was devised using information from previously published series of children with NBR8 ,12–14 and data from a cohort of children with MCD15 (figure 1). This was contributed to by clinicians from Newcastle, Birmingham and Liverpool (NBL algorithm). This algorithm was intended for use in children older than 3 months, presenting to a paediatrician with an NBR, with or without fever. The presence or absence of fever at presentation was not included as it had been shown to have a low sensitivity for MCD.5 ,12 The NICE algorithm was devised using the same previously published series of children with NBR8 ,12–14 but differed by including a much longer list of features suggesting MCD and only taking blood from children with fever, including glucose and lactate in the investigations taken and not advocating antibiotic treatment if leucocyte count was low.16

    Figure 1

    The Newcastle, Birmingham and Liverpool (NBL) algorithm to manage children presenting with non-blanching rash.10 CRP, C-reactive protein; FBC, full blood count; SVC, superior vena cava; WBC, white blood cell count.

    Validation

    Validation was done using three groups of children; retrospective cohorts from three centres, pilot cohorts for the NBL algorithm from selected children in four centres and then a prospective complete cohort from one centre.

    Data were collected from six towns/cities across the UK: four teaching hospitals and three large district general hospitals.

    The NICE algorithm was validated retrospectively using data from all eight cohorts.

    At all participating centres, clinical data were being collected on children who presented with an NBR as part of a local audit.

    MCD was defined as confirmed if there were positive cultures or PCR for Neisseria meningitidis in blood or cerebrospinal fluid, or there were skin scrapings positive for meningococcus (the latter was performed at one centre only). Patients who presented unwell with obvious clinical features of MCD, but who were culture-negative and/or PCR-negative, were defined as probable cases of MCD.17

    Retrospective cohorts

    Three centres had collected retrospective data on children presenting to their hospitals with NBR for audit purposes between 1998 and 2001. These included Royal Hospital for Sick Children, Glasgow, Sunderland Royal Hospital and Ealing Hospital. The NBL algorithm was applied retrospectively to 329 children in these three centres.

    Prospective cohorts

    The NBL algorithm (figure 1) was distributed to a further four centres and pilot data collected prospectively between 2000 and 2002. The algorithm was used by paediatricians in Royal Victoria Infirmary, Newcastle, Newcastle General Hospital and Birmingham Heartlands Hospital. The algorithm was also used in an emergency department in Bristol. While paediatricians were encouraged to use the algorithm, data were only collected by some who did so. These data thus did not include all children presenting with NBR.

    Finally the NBL algorithm was used prospectively for 12 months at Birmingham Heartlands Hospital on all children presenting with NBR between 2001 and 2002. Information on cases of MCD was also actively collected during this time period.

    The data were checked and entered onto a spreadsheet locally and discharge/outcome data were added when those became available.

    NICE algorithm

    The NICE algorithm was applied retrospectively to children from all cohorts.

    Ethical approval was not required, as the algorithms followed best current clinical practice, for both investigation and treatment of MCD. All the data required were recorded as part of routine clinical practice for audit purposes.

    Results

    Data were collected on 625 children, 145 of whom had MCD.

    NBL algorithm

    The NBL algorithm suggested 231 children (37%) should be treated with antibiotics, including all 145 children with confirmed or probable MCD. This gave a sensitivity of 100% and specificity 82%.

    Retrospective cohorts

    The retrospective cohorts comprised 329 cases (median age 2 years; range 1 month–15 years) at three centres, 102 of whom had confirmed or probable MCD, all of whom were identified by the NBL algorithm. The number of cases varied at each centre, as did the percentage treated compared with the percentage with MCD (table 1).

    Table 1

    Validation of NBL and NICE algorithms11 in cohorts of children presenting with non-blanching rash

    Pilot cohorts

    The prospective pilot cohorts included 176 children (median age 2 years; range 2 months–15 years) presenting with NBRs, and the numbers treated are shown in table 1. Overall the NBL algorithm was followed in 143 (84%) of these children. In total 85 children received treatment of whom 26(15%) had confirmed or probable MCD, all of whom were identified by the algorithm. The algorithm was not followed in 28 (16%) children, 17 of whom did not receive treatment despite this being recommended by the algorithm. One of these children had confirmed MCD and did not receive immediate antibiotic treatment. This child was initially thought to have Henoch-Schönlein purpura (HSP) but later developed shock and needed treatment in a paediatric intensive care unit (PICU). Eleven children received treatment despite it not being suggested by the algorithm.

    Prospective cohort

    The prospective cohort included 120 children (median age 4 years; range 0.2–15 years) presenting with NBRs, and the numbers treated are shown in table 1. Overall the NBL algorithm was followed in 89 (74%) of these children. In total 39 (33%) children received treatment of whom 20 (17%) had confirmed or probable MCD, all of whom were identified by the algorithm. The algorithm was not followed in 31 (26%) children, 9 (7%) of whom did not receive treatment despite this being recommended by the algorithm and 7 (6%) received treatment despite it not being suggested by the algorithm. Fifteen children (13%) had HSP. These should have been treated according to the algorithm (purpura), but were not given antibiotics.

    The median ages of those with or without MCD did not differ (2.4 years vs 2.5 years) in either the prospective cohort or the retrospective cohorts.

    NICE algorithm

    The NICE algorithm was assessed using all cohorts retrospectively (table 1). The NICE algorithm suggested 382 children (61%) who should be treated with antibiotics. This included 141 of the 145 children with confirmed or probable MCD. Four children with NBR but no other clinical features of MCD at presentation were not identified by the NICE algorithm; one child had fever but CRP was not measured (although WBC was normal; 7.7×109/L), the second child had no fever and CRP was not measured, the third child had no fever, but a CRP of 146 mg/L; the fourth child had a neutrophil count of 0.89×109/L. All children were given 5–7 days of intravenous antibiotics by their treating paediatricians, although none had MCD confirmed on blood culture and/or PCR. This gave a sensitivity of 97% and specificity 50%; McNemar's test comparing the NBL algorithm with the NICE algorithm showed a statistically significant difference (p<0.001).

    Other bacteraemia

    Two children had bacteraemia due to other organisms (Staphylococcus aureus and Streptococcus pneumoniae). Both had treatment suggested by the NBL and NICE algorithms and both were treated with antibiotics at presentation by clinicians.

    Discussion

    Our study shows that these two algorithms can help paediatricians identify which children with NBR should be treated for probable MCD. In the cohorts studied only one child with MCD did not receive immediate antibiotic treatment, although both algorithms suggested treatment be given, because of the presence of purpura. The algorithms therefore performed as well as, if not better than, the clinicians managing these children. Direct comparison of the two algorithms is not appropriate since data were specifically collected prospectively for the components of the NBL algorithm in more than half the cohorts, but used retrospectively in assessing the NICE algorithm. Both algorithms showed high sensitivity and negative predictive values, although the specificity of the NICE algorithm was lower. This may be because the NICE algorithm includes a much longer list of ‘signs of bacterial meningitis and meningococcal septicaemia’ than the NBL algorithm which only includes meningism, capillary refill >5 s, irritability and lethargy.

    In order to treat those with MCD, it is accepted that the algorithms are broad, and some children are still treated unnecessarily.

    The NBL algorithm correctly identified all children with proven or probable MCD who presented with NBRs. The NICE algorithm identified all children with proven MCD. However four children with probable MCD who were treated would not have been given treatment according to the NICE algorithm. One of these children had a low leucocyte count, which is known to be associated with severe MCD18 but does not trigger antibiotic treatment in the NICE algorithm. Two other children did not have fever on presentation, so would not have had CRP measured in the NICE algorithm (this was markedly raised).

    Raised CRP is a marker of MCD in children presenting with NBR.5 Measuring CRP in all children with NBR (not just those presenting with fever) might help identify more cases of MCD. This could be included in the NICE algorithm in place of lactate and glucose, which have no evidence to support their use in identifying MCD in children with NBR.

    None of the participating centres treated all children with NBR with antibiotics, this varied from 19% to 66%, possibly decreasing over calendar time. While some experts advocate treating all children with NBR for MCD,4 it would mean giving unnecessary antibiotics to 80–90% of children with NBR. Given the recent increases in antibiotic resistance it might be wise to avoid this.9 Many paediatricians only treat children with NBR who have other features suggesting MCD.7 It is vital that those with MCD presenting with NBR are identified and treated. Previous algorithms, such as the ILL criteria, may miss some cases of MCD.10 The NBL algorithm did seem to identify those with MCD across a number of different centres and settings and enabled a directed approach away from universal treatment of NBR, with only 19–66% of children with NBR being given antibiotics. The NICE algorithm suggested treatment more often than the paediatricians managing children in the cohorts.

    Treating all children with NBR for MCD with antibiotics will not treat all those with MCD. NBR is absent in up to 20% of cases of MCD4 and 38% present with a maculopapular rash.19 These presentations can delay the diagnosis of MCD.20

    The NBL algorithm performed well in both retrospective and prospective cohorts, across a variety of settings and during times of variation in the incidence of MCD. The number of children with NBR attending the various centres varied considerably. There were an increased number of cases of MCD in one city (centre 1) during the time of the retrospective audit there. The retrospective cohorts tended to have a higher proportion of cases of MCD. This may be because of difficulties identifying all children who had presented with NBR retrospectively. The pilot cohort had variable numbers of children, since these data were collected when clinicians remembered to record the data. The 1-year prospective cohort is thus the most representative of routine clinical practice. However the NBL algorithm still identified all cases of MCD in every cohort.

    The cohort seen in an emergency department (centre 6) was small, however noticeably fewer children were treated in this cohort. Emergency department doctors may see a different group of children with NBR than are seen by paediatricians and thus have a different threshold for giving antibiotics.7 Further studies of these algorithms in children presenting to emergency departments are thus required before it can be recommended in this setting.

    The algorithms could be improved. While purpura was highly predictive of MCD,21 a significant proportion of children with purpura had HSP. If children with HSP could reliably be identified, then these children would not need to be treated with antibiotics. Validated criteria for confirming HSP are needed in this group. ‘Palpable purpura’ in a child was reported to be highly suggestive of HSP,22 but several authors suggested that this criterion is inadequate. New criteria for the classification of vasculitis in children have been proposed.23 The diagnosis of HSP now requires palpable purpura, plus one of: (1) any renal symptom; (2) IgA deposition in tissue biopsy; (3) arthralgia or arthritis; (4) abdominal pain.23

    However children with MCD can present with a rash resembling HSP with limb and/or abdominal pain,24 leading to delays in treatment.25 The one child with MCD who had delayed treatment in our audit was thought to have HSP because of purpura and joint pain. Robust criteria for differentiating HSP from MCD are thus required. Other features in the algorithm could be removed, without any loss of performance (eg, respiratory rate). Further studies assessing the cut-off value for CRP are also required as different values have been suggested (6 mg/L or 28 mg/L).5 ,26 The use of other biomarkers (such as procalcitonin) may also improve the ability to detect MCD at an early stage.26

    The cohorts used in this study were collated some years ago, when the incidence of MCD was high (especially cohort 1). The incidence of MCD has fallen since then and is likely to decrease further with the introduction of meningococcal B and ACYW vaccines in the UK.27 ,28 However these vaccines will not prevent all cases of MCD, so paediatricians will still need to assess whether children with NBR may have MCD, although the risk of this may be lower in vaccinated children. Validated algorithms to identify which children with NBR should be given antibiotics will thus still be required, although further audit of their use in a low-incidence setting will be needed.

    Children with NBRs may occasionally have bacteraemia caused by organisms other than meningococcus.14 The one child with S. aureus bacteraemia had purpura so was identified by the NBL and NICE algorithms (and treated by the admitting clinicians). The one child with pneumococcal bacteraemia had fever and a raised leucocyte count (33.8×109) so was identified by the NBL and NICE algorithms (and treated by the admitting clinicians). The algorithms thus also identified children with other significant bacteraemia.

    In conclusion, many paediatricians do not want to treat all children presenting with NBRs with intravenous antibiotics. We suggest that these algorithms, when used in conjunction with clinical judgement may help decide which children warrant antibiotics, and would be of use to junior paediatricians who may be increasingly less familiar with MCD.

    References

    Footnotes

    • Presented in part to the Annual Meeting of the Royal College of Paediatrics and Child Health 2004 and 2012.

    • Funding The Johanne Holly Meningitis Fund funded Dr Riordan's initial work.

    • Collaborators We are grateful to the following groups for sharing data from their local audits. The Non-Blanching Rash Audit Group; Sarah Bridges, Southmead Hospital, Bristol; Linda Dykes, Sunderland Royal Hospital; Anu Morjaria, Ealing Hospital; Tina Newton, Birmingham Heartland Hospital; Chris Richards, Royal Victoria Infirmary, Newcastle; Fiona Shackley, Royal Hospital for Sick Children, Glasgow; Ansgar Thimm, Newcastle General Hospital.

    • Contributors FAIR designed the NBL algorithm, analysed the data for validating the NICE algorithms, and drafted and revised the paper. He is the guarantor. JC designed the NBL algorithm and revised the paper. LJ analysed the data for validating the NBL algorithm, and drafted and revised the paper.

    • Competing interests None declared.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data sharing statement Anonymised data on cohorts of children with non-blanching rash—giving clinical and laboratory features in the algorithms are kept securely by FAIR in Liverpool.

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.