• 
• 

Three Part Question

In [adults presenting at an emergency department] can a [triage nurse] accurately predict [hospital admission]?

Clinical Scenario

A busy emergency department has a long list of patients in the waiting area and several patients have recently arrived on trolleys from the ambulance service. There is pressure to get patients seen and either discharged or admitted as soon as possible. You wonder if asking the triage nurses to determine whether they think each patient will need admitted will speed the process along by allowing earlier booking of inpatient beds.

Search Strategy

Database: Embase <1974 to 2016 December 16>

[nurse.mp. or exp nurse/ or exp emergency nurse practitioner/] AND [(emergency care.mp. or exp emergency health service/ or exp emergency care/ or exp emergency ward/) OR (emergency medicine.mp. or exp emergency medicine/)] AND [(hospital admission.mp. or exp hospital admission/) OR (exp hospital bed/ or exp hospitalization/)] AND [(exp prediction/) OR (scoring system.mp. or exp scoring system/) OR (decision support.mp. or exp decision support system/) OR (clinical assessment tool.mp. or exp clinical assessment tool/) OR (exp medical decision making/ or decision making/ or exp clinical decision making/)]

Database: Ovid MEDLINE(R) 1946 to Present with Daily Update

[nurse.mp. or exp Nurses/] AND [(exp Emergency Medicine/ or emergency medicine.mp.) OR (emergency care.mp. or exp Emergency Medical Services/) OR (emergency health service.mp. or exp Emergency Medical Services/)] AND [(exp Hospitalization/ or hospital admission.mp.) OR (hospital bed.mp. or exp "Length of Stay"/) OR (hospitilisation.mp.)] AND [(prediction.mp.) OR (admission prediction.mp. or exp Patient Admission/) OR (scoring tool$.mp.) OR (scoring system.mp.) OR (exp Decision Making/ or exp Decision Support Techniques/ or exp Decision Support Systems, Clinical/ or decision support.mp.) OR (clinical assessment tool.mp.) OR (clinical decision making.mp. or exp Decision Making/ or exp Clinical Decision-Making/)]

Trip Database
Admission prediction triage

Clinical Trials
Triage admission

Titles and abstracts were screened and those thought to be relevant to the three part question were selected. Abstracts were available for all the papers identified, and the search was not limited to English. In addition, those articles selected were further screened by reviewing references and cited articles. Best Bets databases, Google Scholar, Clinical Trials and the grey literature were searched to ensure adequacy of the search strategy.

Search from EMBASE and MEDLINE were combined within Endnote and duplicates were excluded. TRIP does not enable this function and titles were searched online. On the review of titles, 45 abstracts were considered, and 12 papers were identified for full text review. 2 were limited to medical patients only. 1 was on derivation of a tool for predicting admission. 9 papers remained.

Search Outcome

231 papers were found. 9 were relevant and of sufficient quality to be included.

Relevant Paper(s)

Author, date and country	Patient group	Study type (level of evidence)	Outcomes	Key results	Study Weaknesses
Brillman et al 1996 USA	5,106 patients at an academic emergency department with 65,000 patients/year	Prospective crossover study	Nurse prediction of admission	Sensitivity 41.3%, specificity 93.8%, PPV 30.2%, NPV 96.1%	Limited data collection to 07:00-23:00 and excluded CAT 1. Unclear data collection
			Nurse triage categorisation	Not accurate for predicting admission
Kosowksy et al 2001 USA	531 patients at an urban academic emergency department with 75,000 patients/year	Prospective observational study	Nurse prediction of admission using 5 point likert scale	Sensitivity 61.7% (95% CI 51.7-70.8), specificity 90.1% (95% CI 86.7-92.7), PPV 61.7% (95% CI 51.7-70.8), NPV 90.1% (95% CI 86.7-92.7)	Limited data collection to 12:00-20:00. Excluded 'fast tracks' and CAT 1. Small sample size. High drop out.
			Nurse prediction of level of care	Poor PPV for specific level of care
Holdgate et al 2007 Australia	1,342 patients at two tertiary emergency departments with 100,000 patients/year	Prospective study	Nurse prediction of admission	Accuracy 75.7% (95% CI 73.2-78.0), sensitivity 65.1% (95% CI 61.1-69.1), specificity 83.3% (95% CI 80.4-85.9)	High drop out
			Nurse prediction of discharge in patients with injuries	Accuracy 90.9%
			Nurse prediction of admission at extremes of triage score (1 and 5)	Accuracy 94.7% and 100% respectively
Beardsell and Robinson 2010 UK	2,848 patients at an academic emergency department with 85,000 patients/year	Prospective study	Nurses prediction of admission	Sensitivity 67.67% (95% CI 62.76-70.41), specificity 84.79% (95% CI 83.24-86.25), PPV 54.23% (95% CI 50.57-57.85), NPV 90.39% (95% CI 89.05 to 91.62)	Single centre. Limited 2 week data collection.
			Nurses prediction of admission in patients managed in resus or arriving by ambulance	Sensitivities 90.41% (95% CI 4.43 to 94.65) and 78.19% (95% CI 73.27 to 82.59) respectively
			Nurses prediction of admission in self-presentation and in children	Sensitivities 50.84% (95% CI 44.3 to 57.36) and 56.52% (95% CI 46.96 to 65.74) respectively
Peck et al 2012 USA	767 patients in an academic emergency department with 12,762 patients/year	Prospective study	Nurse prediction as one of 3 models to evaluate admission predictions	R-squared value of 0.5243 with an average difference in residuals of 1.87	Single centre. Small yearly census. Limited data collection window. Limited demographic data.
Stover-Baker et al 2012 USA	1,164 patients in a community emergency department with 76,000 patients/year	Prospective study	Nurse prediction of admission using 3 point likert scale	Accuracy 75%, sensitivity 75.6% (95% CI 71.3-79.5), specificity 84.5% (95% CI 83.1-85.8), PPV 62.2% (95% CI 58.7-65.4), NPV 91.1% (95% CI 89.6-92.5)	High drop out. Limited patient group as only self-presenting. Non-consecutive sampling.
Kim et al 2014 Australia	100,123 patients from an urban tertiary emergency department with 74,000 patients/year	Retrospective observational cohort study	Nurse prediction of admission	Accuracy 73%, sensitivity 64.7%, specificity 86.7%	Retrospective record review without prospective validation of models. Single logistic regression used. Not primarily designed to test nursing prediction.
			Admission prediction model (using admission characteristics) vs nurse prediction	ROC area 0.80 vs 0.75 (p <0.001)
			Admission prediction model (using admission characteristics) plus need for blood tests	Accuracy improved from 74% to 76% (p< 0.001)
Cameron et al 2016 UK	1,829 patients in an urban academic emergency department with 86,000 patients/year	Prospective observational study	Nurse prediction of admission	Accuracy 79.0% (95% CI 77.0 to 80.8), sensitivity 81.2% (95% CI 78.2 to 84.0), specificity 77.4% (95% CI 74.8 to 79.9)	Single centre. Limited data collection window.
			Prediction of admission using objective score with nurse veto when clinical certainty ≥95%	Accuracy 82.5% (95% CI 80.7 to 84.2), sensitivity 77.0% (95% CI 73.9 to 80.0), specificity 86.3% (95% CI 84.0 to 88.3)
Alexander et al 2016 Australia	5,135 patients in an urban academic emergency department with 100,000 patients/year	Prospective study	Nurse prediction of admission	Accuracy 83.8%, sensitivity 71.5% (95% CI 68.9-73.9), specificity 88.0% (95% CI 86.8-90.0), PPV 66.9% (95% CI 64.6-69.4), NPV 90.0% (95% CI 89.0-91.0)	High drop out rate. Voluntary data collection. Single centre.

Comment(s)

In the above literature, reported sensitivity for nursing staff predicting admission varies from 41.3% to 81.2%. Each study has limitations based on data collection with variable data reporting. In three of the studies, confidence intervals and raw data were not presented. In addition, several of the studies included were not solely focused on the primary question of this review. Variation in phrasing of the question asked of nurses may have had an influence in changing response - for example admit/discharge vs admission likely/unlikely. Three studies used a likert scale and one a visual analogue scale and attempted to correlate increased confidence with accuracy. The Beardsell and Robinson paper makes the point that the high NPV of nurses may be of more utility in predicting discharge than admission. However, this does not help in the original scenario of this review where the aim was to book beds earlier at triage. It may, on the other hand, provide an opportunity to fast-track low acuity patients for rapid assessment and discharge if appropriate.

Clinical Bottom Line

The reported sensitivity of nursing staff is not sufficient to support early booking of beds at triage. It appears that alternative strategies such as nursing judgement in conjunction with an objective model to predict admission or nursing judgement to predict discharge instead may be more useful.

References

1. Brillman et al Triage: limitations in predicting need for emergent care and hospital admission. Ann Emerg Med 1996 Apr;27(4):493-500
2. Kosowksy et al Can emergency department triage nurses predict patients' dispositions? Am J Emerg Med 2001 Jan;19(1):10-4
3. Holdgate et al Accuracy of triage nurses in predicting patient disposition Emerg Med Australas 2007 Aug;19(4):341-5
4. Beardsell and Robinson Can emergency department nurses performing triage predict the need for admission? Emerg Med J. 2011 Nov;28(11):959-62
5. Peck et al Predicting emergency department inpatient admissions to improve same-day patient flow. Acad Emerg Med. 2012 Sep;19(9):E1045-54
6. Stover-Baker et al Triage nurse prediction of hospital admission J Emerg Nurs. 2012 May;38(3):306-10
7. Kim et al Predicting admission of patients by their presentation to the emergency department Emerg Med Australas. 2014 Aug;26(4):361-7
8. Cameron et al Predicting admission at triage: are nurses better than a simple objective score? Emerg Med J 2016 Feb 10;34(1):2-7
9. Alexander et al Can Triage Nurses Accurately Predict Patient Dispositions in the Emergency Department? J Emerg Nurs. 2016 Nov;42(6):513-518