Febrile children can pose a real challenge to clinicians in the Emergency Department. Identifying and trying to predict those at high risk of severe or invasive bacterial infection is particularly important as there are enormous implications for altering the course of their illness and resource allocation and research initiatives.
There are many clinical scores, but their predictive performance for poor outcomes in undifferentiated febrile children has yet to be discovered.
The 2015 Sepsis-3 definition for adults has been widely discussed in the #FOAMed world. If you want a quick breakdown, then listen to Scott Weingart’s interview with Merv Singer (one of the authors), and read Josh Farkas’ take and the views from the St Emlyn’s crew. We know that patients can have SIRS criteria without sepsis (take my obs after I’ve been to the gym), so we need an alternative method to help rapidly detect those patients suffering from sepsis and the subset with septic shock.
In adults, we can use the SOFA score (Sequential Organ Failure Assessment), which comprises several parameters, but what can we do to help identify children in septic shock?
Let’s take a look at this paper first…
Schlapbach LJ, MacLaren G, Festa M, Alexander J, Erickson S, Beca J, Slater A, Schibler A, Pilcher D, Millar J, Straney L. Prediction of pediatric sepsis mortality within 1 h of intensive care admission. Intensive Care Medicine. 2017 Feb 20:1-2.
Who are researchers?
It was carried out on behalf of the Australian & New Zealand Intensive Care Society (ANZICS) Centre for Outcomes & Resource Evaluation (CORE) and the ANZICS Paediatric Study Group (PSG). The CORE group, according to their website, “…provide audit and analysis of the performance of Australian and New Zealand intensive care.” The Paediatric Study Group collects data from the six Australian PICUs, a single New Zealand PICU, and adult ICUs admitting children.
What sort of trial was it?
This was a multicentre binational cohort study. To remind you – a cohort study takes a group of people, in this case, children, admitted to the PICU/ICU and follows them over time, often looking to see what variables are related to a pre-set outcome.
Who were the population studied?
Patients – All patients under 16 years of age admitted to PICU or a general ICU with a principal diagnosis of either sepsis or septic shock. These patients were compared with the larger group of patients admitted with invasive infections (± septic shock).
Outcome measure – The primary endpoint was 30-day mortality. This data was available for 100% of the patients.
How is paediatric sepsis defined?
The 2005 (!) international pediatric sepsis consensus conference defines severe sepsis as…
Sepsis plus cardiovascular organ dysfunction OR acute respiratory distress syndrome OR two or more other organ dysfunctions
This definition doesn’t exactly roll off of the tongue. Therefore, severe sepsis amounts to hypotension (<5th percentile for age) OR vasopressor use OR hyperlactataemia.
What were the results?
Over the 4 year study period 42,523 patients under the age of 16 were admitted to ICU.
If you look at those deaths related to sepsis/septic shock then there appears to be an error in the manuscript (to me at least, please feel free to correct me). For all patients with sepsis the median time from ICU admission to death was only 1.9 days, with 36.8% dying within 24 hours and 50.7% within 48 hours. The authors then state that in patients with septic shock and no co-morbidities, the median time to death was 16 hours, with 54.5% dying within 24 hours and 72.7% within 48 hours. This seems highly counterintuitive to me. I would expect patients with co-morbidities to die sooner. So why might those children with co-morbidities last longer? The pre-specified co-morbidities include prematurity, congenital heart disease, chronic respiratory disease, chronic neurological disease and immunosuppression.
Mortality was also independently correlated with lactate on presentation to the ICU. A lactate of ≥2, ≥3 or ≥5 mmol/l was associated with an adjusted mortality of 7.4, 8.4 and 9.5%.
What were the authors conclusions?
There is no current gold standard for identifying sepsis, so the authors tried several methods, including multivariate logistic regression, to develop a mortality prediction instrument. To do this, they used data from the time of first face-to-face contact between the patient and a doctor from the ICU or the arrival of the PETS/NETS team to one hour after arrival in the ICU. The researchers found that the following markers were the best predictors of mortality.
With 3 out of 4 previously healthy children dying within 48 hours of admission, early identification of those most at risk of death might be beneficial and help identify those patients in which alternative therapies, such as ECMO, may be useful.
What other PICU scoring systems are out there?
There are a few more scoring systems out there, other than SOFA, that have been shown to have a good correlation with each other. They include:
- PRISM III – Pediatric RiSk of Mortality
- PEMOD – PEdiatric Multiple Organ Dysfunction scoring system
- PELOD – PEdiatric Logistic Organ Dysfunction scoring system
- PIM2 – Pediatric Index of Mortality2
What does this actually mean for me, in practice?
I don’t see many sick septic children in the emergency department. So I don’t think introducing a paediatric sepsis score based on the above variables would make a difference to my practice. However, I hope that if I were looking after a child that had arrested and had fixed, dilated pupils, I would recognise that they were very, very sick indeed.
This study is based on ICU rather than combined ICU/ED data. It would be interesting to know the time frame from the triage to the time to be seen by ICU/transferred and if this impacts mortality.
What about looking at ED data?
Long E, Solan T, Stephens DJ, et al. Febrile children in the Emergency Department: Frequency and predictors of poor outcome. Acta Paediatr. 2020; 00: 1– 10
What was the aim of this study?
This retrospective observational study aimed to determine the frequency of poor outcomes in undifferentiated children presenting to the ED with fever and evaluate predictors of poor outcomes. The authors defined ’poor outcome’ as the development of new organ dysfunction and the requirement for organ support therapy. In addition, they included vital signs, blood tests, and clinical scores as predictor variables.
What was the study design?
This is a retrospective cohort study. It was conducted in the ED in a large tertiary referral centre (single centre study), and full ethical approval was obtained.
Who were the study participants?
All children with ‘fever’ in their triage description or an initial triage temperature of >38.0°C were included, with no exclusion criteria.
How was the study performed?
Data were extracted from electronic medical records. This included demographic data, vital signs, blood test results, diagnosis, disposition, organ support therapies, organ dysfunction scores for patients admitted to PICU and mortality.
To ensure accuracy, one hundred electronic medical records were randomly selected and manually checked.
What were the study team looking for?
The primary outcome of this study was the frequency of new organ dysfunction and the requirement for organ support therapy in the study population, two indicators of severe illness.
The study team examined the following variables to see if any could predict children at risk of poor outcome:
- vital signs: heart rate, respiratory rate, blood pressure, and GCS
- blood tests: venous lactate, creatinine, white cell count, platelet count, and INR
- clinical scores: SIRS, qSOFA, and qPELOD-2
What kind of statistics did they use?
The chart above can be constructive when thinking about statistical analysis. The type of data collected determines the most appropriate means of analysis. This study included both continuous and categorical variables.
For continuous variables, descriptive statistics were used i.e. data was reported using median and inter-quartile ranges.
In this study, continuous variables refer to demographic data such as age, sex, weight, vital signs (temperature, heart rate, blood pressure, respiratory rate, Glasgow coma score) and blood results (including lactate, creatinine, INR, platelet count and white cell count). The use of median and inter-quartile ranges is most appropriate for this data type. The median is the value in the “middle” of the distribution, with 50% of the scores having a value more significant than the median and 50% having a value smaller than the median. The interquartile range (IQR) is the range of values that reside in the middle 50% of the data.
Frequency with percentage was used for categorical variables.
For this study, categorical variables refer to the clinical scores used, i.e. SIRS, qSOFA and qPELOD scores. Describing the data in this way is appropriate as it means the frequency that the data occurred may be expressed as a percentage.
The association between initial vital signs, blood tests, clinical scores and the development of new organ dysfunction and requirement for organ support therapy were reported as odds ratios (OR) with 95% confidence intervals (CI).
Odds ratios are usually used to compare the relative odds of the occurrence of the outcome of interest (e.g. development of new organ dysfunction), given exposure to the variable of interest (e.g. initial vital signs). The OR represents the odds that an outcome will occur given a particular exposure, compared to the odds of the outcome occurring in the absence of that exposure. The confidence interval (CI) is used to estimate the precision of the odds ratio and may be thought of as a way to measure how well your sample represents the population you are studying. A large CI indicates a low level of precision of the OR, whereas a small CI indicates a higher precision of the OR. This study uses 95% confidence intervals which means that there is a 95% probability that the confidence interval will contain the true population mean and in practice, is often used.
The discriminative ability of predictor variables was measured using the area under the receiver operating characteristics curve (AUROC), with sensitivity and specificity calculated for each variable. i.e. vital signs, blood tests and clinical scores.
The Receiver Operating Characteristic (ROC) curve is commonly used in statistics and can be confusing. Put simply, the curve is used to plot sensitivity versus false positive rate for several values of a diagnostic test. It is a graphical measure which illustrates the trade-off between sensitivity and specificity in tests that produce results on a numerical scale, rather than as an absolute positive or negative result. In this study, the AUROC is used to determine the sensitivity and specificity of each of the variables used.
What were the results?
Over the 6-month study period, 6217 (13.8%) children presented to the ED with a febrile illness. This represented just over one-eighth of the overall presentations to the ED. Approximately two-thirds of these children were discharged home (65.4%), a third were admitted to hospital (34.6%), with 0.5% (32 of the 6217 children in the study) admitted to PICU. Slightly more than half of the children, 58.3%, were under 3.
New organ dysfunction was very rare, in (0.4% or 27 children). 10 required organ support therapy (inotropes for 0.2%, mechanical ventilation in 6, renal replacement therapy in 1, and extra-corporeal life support in 1).
The best performing ED predictors of new organ dysfunction were: GCS <11, INR≥ 1.2, lactate ≥ 4.0mmol/L, and qPELOD-2 (SBP) score ≥ 1.
The best performing predictors of the requirement for inotropic support were: initial hypotension using qPELOD 2 (SBP), lactate ≥4mmol/L, INR ≥ 1.2, and qPELOD (SBP) score ≥ 1
The best predictors of the requirement for mechanical ventilation were: GCS <11, lactate ≥4mmol/L, INR ≥ 1.2 and qSOFA=3.
The bottom line
All predictor variables had poor test characteristics for the development of new organ dysfunction and the requirement for organ support therapy.
This is a good study; the results are easy to follow, and, importantly, they meet the study aims. Furthermore, the large sample size gives this study good internal validity, i.e. the extent to which the observed results represent the truth.
Overall, this study supports our clinical experience. Poor outcomes in febrile children are, thankfully, rare. Less than half a per cent of children in this study developed new organ dysfunction—even fewer required organ support therapy. However, the infrequency of these outcomes in the study population means that the use of “predictor variables” is not particularly helpful. A few take-home messages:
Vital signs – Elevated heart rate and respiratory rate were common findings in undifferentiated febrile children. This did not confer an increased risk for the development of organ dysfunction or the requirement for organ support therapy.
Take abnormal GCS seriously though – in this study, very few children had a GCS <11, but when it was low, GCS score was a strong predictor of the requirement for mechanical ventilation.
Blood tests – Remember to check lactate! Elevated venous lactate significantly increased the odds for the development of new organ dysfunction and the requirement for organ support therapy (both mechanical ventilation and inotropic support), with increasing risk the higher the lactate climbed. Elevated initial creatinine and initial INR also signified increasing severity of illness.
Clinical scores – in this study, clinical scores performed variably. They can be helpful but may be more useful in the PICU setting.
The external validity of this study is also strong; the results seem to be generalisable to our own population. Given the lack of exclusion criteria, the results of this study may be applied to any setting where undifferentiated febrile children are cared for.
Were there any limitations to this study?
This is a retrospective, observational, single-centre study using data extracted from an electronic medical record. Retrospective studies may be subject to information bias (by missing information) or selection bias (because individuals are selected after the outcome has occurred). However, this study limited selection bias by including all patients with fever.
In addition, a single-centre study may be limited by using local policies and guidelines rather than disease severity, reducing the external validity/generalisability of the findings.
The outcomes of this study are rare, but the authors attempted to overcome this by using a large sample size of over 6000 children. However, because the outcomes were uncommon, the predictor variables had wide confidence intervals.
Will this study change my practice?
This study is unlikely to change our practice. However, it does provide food for thought. In keeping with our clinical experience, the development of new organ dysfunction and the requirement for organ support therapy is rare among febrile children presenting to the ED.
This study emphasises that predicting poor outcomes in this patient group is difficult. Vital signs, blood tests and clinical scores were poor predictors. This highlights the importance of remaining particularly vigilant concerning undifferentiated febrile children.
A final comment from the authors – Elliott Long and Franz Babl
Thank-you for the opportunity to comment on our article titled ‘Febrile children in the Emergency Department: frequency and predictors of poor outcome’. The associated review covered all of the major aspects of the study
A few additional points that may have been buried in the data:
Though the study was primarily focused on severe infection (sepsis), we included a broader cohort of undifferentiated children with febrile illness presenting to the ED. This was somewhat exploratory, as we suspected that many children would be treated for sepsis (i.e.- admitted to hospital for IV antibiotics and one or more fluid bolus), but would not receive the diagnosis of sepsis. Interestingly, this was borne out in the study findings. The majority of children treated for sepsis did not receive the diagnosis of sepsis. This included the ‘severe end of the spectrum’ of children admitted to ICU; the most common diagnosis in this group of children was ‘acute febrile illness’. We interpreted this finding as being due to the hesitancy of clinicians to label undifferentiated febrile children with the diagnosis of ‘sepsis’ early in their treatment. Prospectively, we all hope kids will ‘turn the corner’ and physiologically improve after basic resuscitative measures… until they don’t! Also, children with more specific diagnoses, such as appendicitis or pneumonia, were more likely to receive these as working (admission) diagnoses even when at the severe end of the spectrum and receiving treatment consistent with sepsis.
The majority of febrile children admitted to ICU did not require (new) organ support. These children included those with meningitis <2months of age, children with croup requiring multiple doses of nebulised adrenaline, children with pneumonia with large pleural effusions, and children on ventricular assist devices. These ICU admissions were based on local policy and procedure, and may not be generalisable to other health services. Studies using ICU admission as an outcome measure should be interpreted with this in mind.
From a ‘big picture’ perspective, this study highlights two major issues for clinicians and researchers when dealing with sepsis.
Clinicians caring for children with febrile illness at different stages of their hospitalisation have different frames on the same disease that we all call sepsis. From an ED and acute care perspective, children with fever are un-differentiated, the majority have a mild, self-limited illness, and can be safely discharged home. The challenge for front line clinicians is early recognition of severe disease- finding the needle in the haystack. From an ICU perspective, children with fever are differentiated, the majority have severe disease and require close monitoring and/or organ support. The challenge in ICU is risk stratification. Understanding these differences in perspective is crucial for communication between clinicians caring for children at different stages of their hospital journey, and for researchers designing studies involving children with sepsis.
As a result of poor outcomes being so rare, interventional trials that aim to capture patients at the entry point of acute care – before they are differentiated – will need to be pragmatic, large, and use composite outcomes. An example of such a study is PROMPT Bolus, which compares 0.9% saline to balanced fluids for sepsis resuscitation and initial maintenance. The study will include pragmatic entry criteria: patients receiving treatment for sepsis (IV antibiotics and >1 fluid bolus). The study will enrol >8000 patients from 3 research networks (PECARN in the United States, PREDICT in Australia / New Zealand, and PERC in Canada), and will use the composite outcome of Major Adverse Kidney Events on day 30 (MAKE30) as the primary outcome. This is probably the model that will be required to answer fundamental questions regarding early sepsis therapies in future.
The NICE screening approach to sepsis
Failing to identify those with “sepsis” has devastating consequences. We’d all love an evidence-based clear-cut path for flagging and managing febrile children at risk of sepsis. Currently, the approach in the UK is predicated on the NICE SEPSIS (NG 51) screening system, which has anecdotally performed poorly with concerns it is poorly specific (i.e. lots of false positives). Nijman and colleagues aimed to objectively assess the impact of the NICE Sepsis screening approach in children.
Nijman RG, Jorgensen R, Levin M, Herberg J and Maconochie IK. Management of Children With Fever at Risk for Paediatric Sepsis: A Prospective Study in Paediatric Emergency Care. Frontiers in Pediatric Care 2020; 8:548154. doi: 10.3389/fped.2020.548154
The lead authors looked at the various warning signs of serious infections in febrile children presenting to PED. In addition, they aimed to determine these children’s risk of sepsis and evaluate their subsequent management.
Who did they study?
Over 5000 children (5156 to be exact) aged 1 month to 16 years old presenting with fever over a period of 9 months from June 2014–March 2015 in a single PED at St Mary’s Hospital, UK were analysed. Febrile children with no warning signs of sepsis were then excluded from the final cohort. The second largest group excluded from the final cohort was children with a complex medical history (n=119). The decision to exclude this particular cohort is important given that ‘complex medical patients’ are more likely to have sepsis. The authors make the valid point that this group has features very different from the intended cohort, such as having different management plans in the context of fever. After these exclusions, plus a few further exclusions (lack of consent, lack of complete data or excluded because the child didn’t have any warning signs) the final cohort was of 1551 children.
What did they do?
They first examined the numbers of febrile children with tachycardia and tachypnea using APLS and NICE (the National Institute of Healthcare Excellence) thresholds. Subsequently, they looked at the numbers of febrile children fulfilling sepsis criteria using well-known sepsis screening tools (NICE traffic light guidelines, SIRS, qSOFA, Sepsis Trust UK trigger criteria).
All the data for this study (vital signs, clinical signs and symptoms, tests, working diagnosis, need for hospital admission, timeliness of interventions) were collected electronically, having been recorded prospectively for all febrile children.
What did they look for?
As a primary outcome the study determined:
- The incidence of febrile children who present with warning signs of sepsis
- How often these children fulfilled paediatric sepsis criteria
- How frequent invasive bacterial infections (IBIs) occurred in this population
- How frequent PICU admissions occurred in this population.
Secondary outcomes included clinicians’ compliance with the paediatric sepsis six care bundle (PS6), what clinical interventions were and were not used from this care bundle and the timeliness of the interventions undertaken.
What did they find?
Almost a third of children aged 1 month to 16 years who presented to the PED had fever (28% to be exact).
41% of these febrile children had one or more warning signs (our study population).
The incidence of IBI was 0.39%. Of these children, only 0.3% required PICU admission.
This meant that 256 children would need to be treated using the sepsis guideline recommendations to catch one IBI. Another way of saying this is the number needed to treat was 256. NNT for any serious outcome was 141.
How did the sepsis guidelines fare?
The thresholds for tachycardia and tachypnoea yielded a high false positive rate.
Adding sepsis criteria to predict the presence of a serious bacterial infection (SBI), IBI or PICU admission was also unreliable, with a lot of false positives.
Lactate levels were not significantly associated with the decision to give IV fluid bolus or presence of SBI, IBI or PICU admission. There WAS, however, a significant association between lactate levels and hospital admission.
Looking at the Paediatric Sepsis 6 Interventions, although many children triggered, two-thirds (65%) of the children with PS6 warning signs had none of the PS6 interventions. And when it came to the ‘golden hour? Only a third (36%) of children with IBI or PICU admission received all PS6 interventions in the ‘golden hour, with only 39 children (2%) receiving a fluid bolus.
What does this all mean?
It is important to note that this study was only conducted in one single PED and in a period that was before the NICE sepsis guidelines were formally implemented into practice. The data was collected for this study via an electronic interface. While large amounts of data can be collected rapidly, there can sometimes be gaps, either due to extraction issues or brevity on behalf of clinicians that don’t give a comprehensive picture.
Data were also only taken from initial triage and not from any clinical deterioration in the ED. Given that acuity changes over time, especially in children with fever, this may have missed subsequent clinical change. However, it is a pragmatic approach given that sepsis screening tools are applied in nearly all Emergency Departments.
The numbers needed to treat were exceptionally high. Despite the allure of a protocol-based screening and management pathway, the benefits of catching true sepsis early must be weighed against the possible unwanted effects of overtreating or overdiagnosing mostly healthy children in a potentially resource-stretched PED. The study does highlight the difficulties we face when screening for a septic child in a generally healthy cohort, the ‘needle in a haystack’.
Essentially, this study shows us that serious infections are rare, and most children categorised as at risk of sepsis’ can be managed conservatively with little intervention other than observation. Our current guidelines have very poor specificity. While they tell us to investigate and treat many children, we as clinicians often rely on our clinical judgement and essentially ‘do nothing’.
Observation and good clear red flagging must not be underestimated. Instead of continuing to research more and better early predictors of sepsis, such as point-of-care biomarkers, perhaps we should look at this from another angle. The lens’s focus can also be flipped; we also need more research on how it can be safe NOT to do anything.
What do the authors think?
The Infections in Children in the Emergency Department (ICED) study is a single centre, prospective observational study. The study describes unique and carefully curated clinical data of febrile children with warning signs of sepsis, from a period prior to the implementation of the NICE sepsis guidelines.
Our results confirm what many paediatricians dealing with acutely unwell febrile children already suspected: that many febrile children have warning signs of sepsis, but that the large majority have non-life threatening infections.
Our findings will hopefully contribute to ongoing discussions about the use of sepsis screening tools in paediatric emergency medicine. Our study makes it clear that current tools lead to a high number of false positive cases, and their usefulness in routine clinical care in paediatric emergency medicine should be questioned. Escalation to senior decision makers of all children with warning signs of sepsis should be aspired, but is seldomly feasible in clinical practice and with unproven impact on reducing missed cases and optimising clinical care for the total cohort of febrile children.
Although all children with serious infections would have been detected by the various sepsis tools, it is now evident that we need better tools to more selectively identify children at the highest risk of sepsis. Future studies should explore the utility of machine learning as well as the potential of combining clinical signs and symptoms with point of care biomarkers.
Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, Bellomo R, Bernard GR, Chiche JD, Coopersmith CM, Hotchkiss RS. The third international consensus definitions for sepsis and septic shock (sepsis-3). Jama. 2016 Feb 23;315(8):801-10.
Goldstein B, Giroir B, Randolph A. International pediatric sepsis consensus conference: definitions for sepsis and organ dysfunction in pediatrics. Pediatric critical care medicine. 2005 Jan 1;6(1):2-8.
Rady HI, Mohamed SA, Mohssen NA, ElBaz M. Application of different scoring systems and their value in pediatric intensive care unit. Egyptian Pediatric Association Gazette. 2014 Dec 31;62(3):59-64.
Ha EJ, Kim S, Jin HS, Bae KW, Lim HJ, Seo JJ, Park SJ. Early changes in SOFA score as a prognostic factor in pediatric oncology patients requiring mechanical ventilatory support. J Pediatr Hematol Oncol. 2010 Nov;32(8):e308-13. doi: 10.1097/MPH.0b013e3181e51338. PubMed PMID: 20818274.