With good intent, a number of clinicians have attempted to develop clinical classification tools to assist in early identification of serious illnesses in children. There have been many such that have spanned a wide range of illnesses. Most clinicians are familiar with asthma scores and croup scores, each intended to quantify the level of respiratory compromise and to imply relative need for more intensive management. As well, most clinicians have heard of clinical tools for evaluation and management of fever in young infants. Each of these tools has its advocates and detractors. Few have been shown to have consistent validity and reliability when applied in day-to-day practice settings. Those that have been shown to be reliable require meticulous attention to detail with exact application of its components, which can make compliance quite challenging.
Perhaps from this perspective, investigators in the United Kingdom conducted a realistic test of applicability and reliability of several Pediatric Early Warning Scores (PEWS) that have been advocated for use in Pediatric Emergency Departments (PED). PEWS are mostly physiology-based scoring systems that are developed to identify children who are in-patients and are at risk for clinical deterioration. Identification of a reliable early warning tool could greatly improve patient safety, by decreasing the number of children with potentially serious illnesses that escape the notice of the clinician(s).
Seiger and colleagues conducted a thorough review of the literature and identified ten PEWS that were pediatric-specific, and methodologically sound. They applied these tools to a cohort of children who presented to their emergency department from 2006 through 2009, utilizing physiological data that was measured and recorded by pediatric ED nurses per their usual triage practices. Missing vital signs were subsequently mathematically imputed using a multiple imputation model, meaning that a missing piece of data was replaced by a value drawn from an estimate of the distribution of the variable.
There were two primary outcome variables: ICU admission and admission to the hospital. ICU admission yielded the best performance of the PEWs tools. For ICU admission, sensitivity ranged from 61% to 94%, and specificity from 25% to 86%. The PEWS with the greatest sensitivity for ICU admission (94.4%) had the lowest specificity (25.2%), and the PEWS with the lowest sensitivity (61.3%) had the greatest specificity (86.7%). For hospitalization, sensitivity ranged from 36% to 85%, and specificity from 27% to 90%. The PEWS with the greatest sensitivity for hospital admission (85.7%) had the lowest specificity (27.1%), and the PEWS with the lowest sensitivity (36.4%) had the greatest specificity (90.5%). Based on these data, the authors suggest that while current PEWS tools can help identify PED patients who might require ICU services, none are both sensitive and specific in this regard, and all lack the sensitivity to be relied upon entirely. Furthermore, of the ten tools tested, only three performed better than a coin toss with regard to indication of need for in-patient management.
The authors commented on the limitations of their study design. One that seems to beg greater emphasis is the large number of missing vital signs measurements, which nine of the ten PEWS tools tested require. Of the 17,943 children in the study cohort that was analyzed, measurements of heart rate, respiratory rate, blood pressure, and oxygen saturation were available in only 51%, 37%, 20%, and 27%, respectively. One can only speculate about the difference in results that might have been, if true physiological measures replaced imputed data in this study.
For many long-time practitioners of pediatric emergency medicine, the results and conclusions of this investigation are likely not surprising. For decades, clinicians have attempted to develop streamlined assessment tools that can be easily applied and yield consistent and reliable results. To the latter investigators can sometimes come close, but those successes invariably come at the expense of ease of application and practical compliance. In this study, the tool with the simplest design did have the single-highest specificities for either ICU admission or hospitalization, but it also had the single-lowest sensitivities for each. Conversely, PEWS tools with the highest sensitivities (and lowest specificities) for both ICU admission and hospitalization were among the most detailed and multi-component.
There is one final observation (of mine) that as a curiosity, begs mention. Each of the PEWS tools identified work of breathing, skin perfusion, level of consciousness, and (all but one) vital signs as the important components of clinical assessment of well-being. This is either reassuring, as all investigators agreed that these parameters are most important; or, this is misleading, as none of the PEWS tools performed at a level that would earn unconditional endorsement. Food for thought.