Article Text

Download PDFPDF

Can emergency department nurses performing triage predict the need for admission?
  1. Iain Beardsell,
  2. Sarah Robinson
  1. Department of Emergency Medicine, Southampton University Hospitals NHS Trust, Southampton, UK
  1. Correspondence to Dr Iain Beardsell, Department of Emergency Medicine, Southampton University Hospitals NHS Trust, Tremona Road, Southampton SO16 6YD, UK; iain.beardsell{at}suht.swest.nhs.uk

Footnotes

  • Competing interests None.

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

View Full Text

Statistics from Altmetric.com

Introduction

In most UK emergency departments (EDs), patients who present for evaluation and treatment undergo a brief initial assessment to ascertain their immediate clinical needs. This is used to prioritise care and direct resources within the department. At this assessment, patients (or others such as ambulance crew) give a brief history to the triage nurse and a triage category is then assigned to the patient. At our institution this process is computerised and includes the Manchester Triage Scale1 and several other mandatory questions (such as allergies and tetanus and immunisation status) and an area for free text.

Since the publication of ‘Reforming Emergency Care’ by the Department of Health, EDs in the UK have been encouraged to ensure shorter waits and it stipulated that, by the end of 2004, no patient should spend more than 4 h from arrival in the ED to admission, transfer or discharge.2 The demands of this target were reduced slightly by the introduction of clinical exceptions to waiting times3 and the lowering of the 100% target to 98% in March 2005.4 This has been further lowered in 2010 to 95% to allow for more clinical exceptions. However, this target presents a continual challenge to EDs and many are struggling to achieve the standard.5

We hypothesised that nurses performing triage may be able to accurately predict patient admission at this first encounter within the ED, prior to any formal medical assessment enabling earlier inpatient bed requests, thereby reducing length of stay within the ED.

Previous studies investigating the ability of nurses at triage to predict the disposition of patients from emergency departments have reported mixed results. A 2-week observational study based in two urban Australian teaching hospitals found the triage nurse predicted the correct disposition in 75.7% of patients.6 The proportions of patients admitted in these units were higher than are seen in UK EDs (36%), with fewer patients seen per annum (51 000 and 49 000) than in our institution which sees >85 000 patients per year. A large proportion of patients were excluded from this study (769/2111 patients, 36.43%).

A similar study performed in the USA demonstrated a positive predictive value of 61.7. The study had relatively few subjects. Of 987 patients only 521 were analysed, all ‘fast-track’ patients (normal vital signs or minor complaints) were excluded and patients were only recruited during an 8 h period of each day.7

A larger study, again from the USA, analysed over 5000 patients, finding a positive predictive value of only 30.2% in triage nurse prediction of admission.8 Critically ill or sick patients or attendances overnight were excluded and admissions were only 6.2%, significantly less than most ED populations. The studies to date have all had significant limitations, restricting their extrapolation to the UK population.

Our primary outcome was to assess the ability of nurses performing triage to predict the need for admission. Secondary outcomes were the accuracy of this assessment in various different patient groups. The following groups were determined a priori having been deemed to have been clinically and operationally useful: those arriving by ambulance; adults conveyed by ambulance; patients self-presenting to the ED; children (aged ≤16 years); adults (aged >16 years) and patients taken immediately to the resuscitation room.

Methods

We conducted a prospective observational study in Southampton University Hospitals NHS Trust, a tertiary-referral teaching hospital. Over a 2-week period (16–30 June 2009) we asked the nurses performing triage in the ED to record, at the end of the triage process, whether they thought the patient would require admission or be discharged. This was facilitated using the departmental patient management system (Ascribe Symphony). The question ‘At the end of the triage make your best guess as to the patient's outcome: Admit—will be admitted to an inpatient ward; Discharge—will be discharged from ED’ was asked by the system and its completion mandatory—that is, triage could not be completed unless this question had been completed and one of the drop-down options (Admit or Discharge) selected. Exclusions were patients who were discharged immediately following triage; redirected to another healthcare provider after triage; self-discharged before completion of the care episode; died in the ED; and those who had been referred to inpatient teams by primary care practitioners. Patient outcomes (admission or discharge) were collected using the Symphony system. Although the nurses involved were aware that this additional question was part of a study, they were unaware of the study hypothesis. Clinicians involved in the patient's care were blinded to the triage nurse's prediction and no new action was taken as a result of it (ie, beds were not booked prior to medical evaluation).

We calculated results for the entire cohort as well as performing a priori defined subgroup analysis of different patient groups where, had this been successful, we could practically implement prediction of admission by nurses at triage following the study.

The following subgroup analysis was performed:

  • Patients arriving by ambulance.

  • Patients self-presenting to the ED.

  • Paediatric patients (age ≤16 years).

  • Adult patients (aged >16 years).

  • Adult patients arriving by ambulance.

  • Patients presenting to the resuscitation area arriving by ambulance.

The patient episodes were analysed and 2×2 tables constructed. The results were analysed using MedCalc.9 The outcome from the ED (either admission or discharge) was regarded as the gold standard. From this, sensitivity, specificity, positive predictive values (PPV), negative predictive values (NPV) and likelihood ratios (LR) and their 95% CIs were generated.

All nurses undertaking triage at our institution have undergone formal teaching and assessment in the use of the Manchester Triage System.1 During the study period there was a wide range of nursing experience of those performing triage, ranging from 1 year of emergency nursing to some senior nurses with >25 years of working in EDs. The seniority of nurses performing triage was not analysed as part of the study. For this predictive tool to be used at our institution, it would have to be consistently sufficiently accurate at all times, regardless of which nurse was performing triage.

Results

During the trial period 3144 patients attended the ED, of which 296 were excluded from the study according to the set criteria, leaving a study cohort of 2848 patients. Within the trial group 1467 were male (51.5%) and 1381 female (48.5%). The mean age was 36 years, median age 32 years (range 1 month to 100 years). The percentage of patients admitted from the study cohort was 21.28% and varied in the subgroups with a high proportion of patients presenting to the resuscitation area being admitted (73.74%) and a lower proportion being admitted after self-presenting (12.14%). This percentage of patients admitted is consistent with those usually seen at our institution (table 1).

Table 1

Numbers of patients admitted and discharged in study period and whole year

Throughout the study we regarded the ability of the nurse performing triage to predict admission as a diagnostic test (table 2). During the study period the sensitivity and specificity of the prediction at triage were generally poor and insufficient to ‘rule in’ or ‘rule out’ the need for an inpatient bed. Sensitivity was particularly poor, with only the small resuscitation subgroup exceeding 90, reflecting the high number of true positives and the low number of false negatives, but in this group the proportion of patients admitted (pretest probability) was also high (74.76%).

Table 2

2×2 table data

The positive predictive value (ie, the post-test probability for our patient cohort) was also poor; it was 54.23% in the whole cohort and only 40.47% in those self-presenting to the ED. This would translate into admission being correctly predicted in the whole cohort for only just over half of the patients (table 3).

Table 3

Sensitivity, specificity, positive and negative predictive values (PPV and NPV) and positive and negative likelihood ratios (LR+ and LR−), with 95% CIs

The LRs calculated allow other institutions with different admission rates to calculate the post-test probability of nurses at triage correctly predicting admission. A positive likelihood ratio (LR+) is generally regarded as somewhat useful if >5 and very useful if >10—that is, a positive result will affect the pretest probability (in this case, the percentage of patients admitted) sufficiently to predict more accurately the need for admission. A negative likelihood ratio (LR−) is somewhat useful if <0.2 and very useful if <0.1 at predicting the patient being discharged.

In our study the LR+ was >5 in only one subgroup (children) and no LR− was <0.2.

Discussion

We have regarded the ability of the nurse performing triage to predict the need for admission as a diagnostic test. Where the proportion of patients admitted and discharged is known (the pretest probability), the PPV and NPV demonstrate the accuracy of that prediction.

The key determinant of whether this policy would be successful is measured by the PPV—that is, if a nurse predicts a patient will be admitted (and therefore a bed prepared and reserved for them on the inpatient ward), how often will this be correct? To put it another way, what is the post-test probability? In our cohort the results for this were poor. For the whole group of patients the PPV was only 54.23%—that is, if the nurse performing triage predicted that the patient would be admitted, they were only correct half of the time.

If the nurse performing triage predicted that a patient would be discharged, the results were improved with the nurse being correct 91% of the time, but this result may in part be due to the large numbers in the cohort who were discharged. It would also still mean that one in 10 patients that they predicted would be discharged would require admission.

At our institution, if a patient was predicted as needing admission, this would have triggered a process where that patient had a named bed reserved for them on the inpatient ward. Over the entire 2-week period, if we had used the prediction by triage nurses as a trigger for booking beds, 745 inpatient beds would have been reserved of which only 404 would have been required, while a further 202 patients requiring admission would not have been predicted. As these decisions happen concurrently, this may mean patients for whom admission had not been predicted waiting even longer than previously, as capacity was reserved for many who did not need it. If this prediction had been used purely to indicate the number of inpatient beds likely to be needed for emergency admissions, the significant overbooking could have led to the cancellation of elective operations and early discharge of patients to free capacity that was not required.

We had hoped prior to the study that nurses performing triage would be accurate enough at predicting the need for admission to aid our inpatient bed management strategy and patient flow through the ED. We were somewhat surprised at the inaccuracy in predicting admission at triage. After conducting this study we would suggest that deciding whether inpatient care is required is a complex mix of clinician experience, wider information gathered from patients, their relatives, carers or primary care physicians and the need for further testing and treatment. The need for admission cannot be predicted on the basis of the limited information gathered at a brief assessment such as triage, even when using structured tools such as the Manchester Triage Scale.

Limitations

These results can only be applied to our own institution with any degree of certainty. Factors that could affect application in other hospitals include different patient populations, thresholds for admission and experience of triage nurses. However, we believe that our hospital is typical of most EDs in the UK.

Pragmatically, we regarded our gold standard as whether a patient was admitted or discharged, not whether they were appropriately admitted or discharged, as this is the outcome that we are concerned with on a day-to-day basis.

Conclusion

These results suggest that nurses at our institution performing triage are inaccurate at predicting who will need admission. We would postulate that the reasons for this are multiple, not least the lack of information available to the nurse at the time of triage.

This study highlights the fact that whether a patient will be admitted or discharged following clinical assessment in the ED cannot be accurately predicted after triage alone, and suggests that it can only be determined after a comprehensive clinical investigation. We would suggest that this includes an appropriate history, examination and testing, and patients cannot be ‘signposted’ to inpatient areas at presentation without a significant number of inappropriate admissions and subsequent wasting of precious resources.

References

View Abstract

Footnotes

  • Competing interests None.

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

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.

Linked Articles

  • Primary survey
    Jonathan Wyatt