Discussion
In this study, we examined the validity of the paediatric component of SATS-N version 4.0 used on patients aged 0–14 years who were transported by ambulance to hospital for emergency medical assistance. We found that only half of the patients were assigned the same triage category in the pre-hospital and A&E settings. The SATS-N form used in both settings had high sensitivity (96 % and 88 %), low specificity (46 % and 60 %) and a very high negative predictive value (99 % and 98 %) for identifying children who needed to be transferred directly from A&E to a high dependency unit. The overtriage rate was low (below 20 %).
In a busy A&E, the need for further triage in addition to that in the pre-hospital setting could be questioned. Being mistakenly assigned a low triage category will have the most serious repercussions for the patient. Although only half (48 %) of the patients retained the same triage category (colour code), we found that the majority (69 %) retained the same triage level (high or low) following triage in A&E, while a small proportion (11 %) were upgraded from low to high. The change in triage level may be due to a change in clinical status or effect of initiated treatment. The difference may also depend on the healthcare personnel's clinical experience in assessing children. The results of this study suggest that children should be triaged in A&E even if this has been done in the pre-hospital setting.
The sensitivity and specificity in the SATS-N form in this study were similar to those observed for other triage tools, although direct comparisons are difficult due to the different nature of health services and methods for measuring validity (11–13) . In a systematic review from 2019, two widely used triage tools were examined: the Manchester Triage Scale and the Canadian Triage and Acuity Scale. Sensitivity was defined as a high triage level in A&E and subsequent admission to the intensive care unit, while specificity was defined as a low triage level in A&E and subsequent outpatient treatment. The sensitivity in paediatric triage tools ranged from 71 % to 93 % and specificity from 69 % to 96 %. The study also showed that triage tools yielded different outcomes across studies, depending on the study population and design, and how the true triage level was defined (11) .
A systematic review of paediatric triage tools found that the validity of various tools was higher in studies conducted in or near the country where the triage tool was originally developed (12) . The paediatric version of SATS used in South Africa has previously proven to be a robust tool, with sensitivity of 91 %, specificity of 54 % and a negative predictive value of 95 % for identifying children who need to be admitted to hospital (14) . In our study of SATS-N, we found roughly the same values, but specificity in SATS-N was slightly lower and the negative predictive value was slightly higher. The varying definitions of a true high triage level and the differences in patient population in terms of the range of medical problems and economic conditions both make direct comparisons challenging.
An ideal triage tool should have low overtriage and undertriage rates. A high overtriage rate can lead to fatigue among healthcare personnel and misuse of resources (9, 15) , and at worst could compromise the care of the most urgent patients (12) . However, it is generally acceptable to have a slightly higher overtriage rate and higher resource utilisation than a high undertriage rate, which can delay the treatment of serious illness. The study from South Africa found that the paediatric version of SATS resulted in an overtriage rate of 45.5 % (14) . There is no consensus on what constitutes an acceptable level of overtriage or undertriage for triage tools. The international trauma community recommends an overtriage limit of around 20–25 % and an undertriage limit of around 5 % (16, 17) . Based on these criteria, SATS-N showed a low overtriage rate (19 % and 18 %) for children transported to hospital by ambulance, an excessive undertriage rate in A&E (12 %), and a satisfactory undertriage rate in the pre-hospital setting (4 %).
The negative predictive value is an important measure of validity in the evaluation of triage tools. In this study, we found that SATS-N had a very high negative predictive value (98–99 %), meaning that patients given a low triage level were highly unlikely to be transferred from A&E to a high dependency unit. This is reassuring in a clinical context. One of the weaknesses of SATS-N was the low specificity, of just 46–60 %. Consequently, a substantial number of patients can be assigned a high triage level that, according to the definition criteria, is incorrect. However, positive and negative predictive values depend on the prevalence of the variable to be identified in a population, and the result can therefore vary across different patient populations.
Strengths and weaknesses
Strengths and weaknesses
This study of the use of SATS-N in a pre-hospital setting aims to expand the knowledge base for the triage system used in the Western Norway Regional Health Authority. The study spans several seasons to encompass the heightened influx of paediatric patients during the winter, which is the peak season for respiratory infections.
A weakness of the study was the inclusion of patients during the ongoing COVID-19 pandemic. During the period with extensive infection control measures in spring 2020, there were fewer acute hospital admissions than usual for somatic conditions for both children and adults (18) . However, several countries reported delayed access to emergency medical assistance for children (19, 20) . The range of illnesses and their severity in patients transported to the hospital during the inclusion period may have been different from that normally observed. Another weakness is that pre-hospital triage levels were not documented for around 10 % of the ambulance patients. This may be related to the prioritisation of infection control measures over documentation of urgency. It could also be associated with the relatively high proportion of short journeys (less than 15 minutes). At the start of the COVID-19 pandemic, the threshold for admission to an intensive care unit may have been higher than usual to avoid overcrowding. However, in our experience, children were given a very high priority by the intensive care unit at Haukeland University Hospital, and we do not believe that this has impacted on the results in this study.
Other weaknesses of the study were the inclusion of patients from only one hospital and one geographical area, and the fact that A&E nurses were not blinded to pre-hospital triage levels and may have been influenced by this information. There is no consensus on which method should be used to validate triage tools or on the definitions of overtriage and undertriage (7, 21) . Our benchmark for a true high triage level, involving direct transfer from A&E to a high dependency unit, may be inaccurate, as healthcare personnel may assign this to a patient as a precaution, even if the patient is not necessarily critically ill. Definitions also do not encompass all conditions with a high urgency level. For example, children with a reduced level of consciousness in connection with a febrile seizure or hypoglycaemia often recover quickly and may not necessarily need to be transferred to a high dependency unit.