a b s t r a c t

Objective: The objective was to determine the accuracy of the outcome predictive scores (modified early warning score [MEWS]; Hypotension, Low Oxygen Saturation, Low Temperature, Abnormal ECG, Loss of Independence [HOTEL] score; and Simple Clinical score [SCS]) in predicting en-route complications during interfacility trans- port (IFT) in emergency department.

Design: This was a retrospective cohort study.

Methods: All IFT cases by ambulances with either nurse-led or physician-led escort, occurring between 1 January 2011 and 31 December 2012, were included. Obstetric and Pediatric cases (age b 18 years) were excluded. The condition of patients was quantified by using the predictive scores (MEWS, HOTEL, and SCS) at triage station and on ambulance departure. The accuracy of predictive scores was compared by the receiver operating charac- teristic (ROC) curves.

Results: A total of 659 cases were included. Seventeen cases had en-route complications (2.6%). The complication rate in physician-escorted transport (2.2%) was similar to that in nurse-escorted transport (2.6%). None of the 57 intubated cases had en-route complications. The area under the ROC curve for MEWS was 0.662 (triage) and

0.479 (departure). The accuracy of MEWS at triage was better than that at departure (P = .049). The area under the ROC curve for HOTEL was 0.613 (triage) and 0.597 (departure), and that for SCS was 0.6 (triage) and 0.568 (departure). In general, the predictive scores at triage were better than those on departure. Conclusion: None of the scores had good accuracy in prediction of en-route complications during IFT. MEWS at triage was among the best one already but was not ideal.

(C) 2016

Introduction

The demand for interfacility transport (IFT) is increasing because of regionalization of health care system in recent years. The decision to transfer is complex and depends on many factors including the capabil- ities of the presenting hospital, capacity at the receiving hospital, and fi- nancial and geographic factors. Most transfers are initiated because of condition-specific recommendations such as care for patients with major trauma, stroke, or acute myocardial infarction, conditions in which the patient requires a higher level of care or specialized service not available in the current hospital, and unavailability of beds [1,2].

Interfacility transport is not without risk. The transport is initiated when the benefits to the patient outweigh the risks of the transport. These risks include clinical deterioration, limited resource availability during transfer, transportation risks, time delay in time-sensitive care,

* Corresponding author at: Department of Accident and Emergency Medicine, Tuen Mun Hospital, 23 Tsing Chung Koon Rd, Tuen Mun, New Territories, HKSAR. Tel.: +852 24685200; fax: +852 24569186.

E-mail address: [email protected] (C.T. Lui).

¹ Dr YK Wong and Dr CT Lui had equivalent contribution to the article and would be listed as co-first authors.

poor communication between facilities, inexperienced personnel, and neglect of patient’s preferences [1]. A recent study found that the com- plication rate of land-based IFT was 6.5% [5]. To minimize the transpor- tation risks, it is important to anticipate the potential en-route complications so as to arrange suitable accompanying personnel and equipment to optimize patients’ outcomes. In this situation, an effective clinical predictive scoring system is needed.

There are numerous predictive scoring systems developed in previ- ous studies for risk stratification and outcome prediction of emergency patients, for example, the Modified early warning score ; Hy- potension, Low Oxygen Saturation, Low Temperature, Abnormal ECG, Loss of Independence (HOTEL) score; Simple Clinical Score (SCS); Sim- plified Therapeutic Intervention Scoring System (TISS-28); Acute Phys- iology and Chronic Health Evaluation Score; Simplified Acute Physiology Score; and Risk Score for Transport Patients. However, these scores were designed for risk stratification of patients in the set- ting of the emergency department (ED) and in-patient and critical care unit, and the outcomes predicted were mortality or surrogate out- comes of in-hospital deterioration. There is no dedicated predictive

http://dx.doi.org/10.1016/j.ajem.2016.02.009

0735-6757/(C) 2016

score for determination of en-route complications during IFT [6]. This study evaluated MEWS, HOTEL, and SCS in predicting en-route compli- cations during IFT departed from the ED.

Methods

Study design and setting

This is a retrospective cohort study. We compared the accuracy of various clinical predictive scoring systems based on bedside clinical pa- rameters and chronic health status on predicting en-route complica- tions during IFT in ED.

The study took place in the Accident & Emergency Department of Pok Oi Hospital (POH) in Hong Kong. In 2012, POH was a 392-bed local hospital in the New Territories West Cluster of Hong Kong, which served a population of more than a million. The ED had an annual attendance of 129,814 patients in 2012. It provided clinical services in- cluding 24-hour emergency service, acute medical service, as well as the Combined Coronary and Intensive Care Unit, which provided critical care and emergency cardiac care, except for those who needed surgical interventions or pediatric patients. There was no emergency obstetric, gynecological, and surgical service in POH. In this cluster, the emergency operations are provided in Tuen Mun Hospital, which is a referral center with support of various specialties. Surgical patients are initially stabi- lized and transferred to Tuen Mun Hospital for further care. All mental patients who require admission would be admitted to the local short- stay ward for psychiatric assessment without primary transport.

For those patients who require IFT, the escort personnel could be para- medics (ambulance crew), a nurse, or the attending emergency physician. There were local departmental recommendations on the escort person- nel. All patients with artificial airways and those hemodynamically unsta- ble or critically ill with life-sustaining devices would require a physician to escort. All obstetric cases in active labor, unless a nurse with midwifery skills is available, would require the attending physician to escort to labor ward. Those cases which were considered to have a higher risk of complication or required nursing care would be escorted by nurses. All transports were carried out on ground by ambulances.

Data collection

All IFT cases with either nurse-led or physician-led escort that de- parted from the ED between 1 January 2011 and 31 December 2012 were included. Eligible cases were retrieved from the local interfacility transfer registry which included all Interhospital transfer data. Obstetric and pediatric cases (age b 18 years) were excluded. All included cases were transported by ambulances. Those patients who had been admit- ted to wards in local hospital before secondary transport were excluded. Information collected included demographic data, physiological sta- tus, preexisting disease, chronic health condition, and en-route physio- logical deterioration. Vital signs, including blood pressure, heart rate, respiratory rate, consciousness level and Glasgow Coma Scale (GCS), at both triage and just before departure from ED were retrieved. Rele- vant variables of premorbid conditions and the information about the present illness, which were components of the scores, were retrieved and presented. These included whether the patient was intoxicated, whether the patient was a trauma victim, the ambulatory status, a nurs- ing home resident, history of diabetes, present illness of new-onset stroke, abnormal ECG results, and breathlessness on presentation. Ab- normal ECG result was defined as any ECG abnormalities that were iden- tified by the algorithm of the ECG machine, excluding simple Sinus bradycardia and tachycardia. New-onset stroke was defined as new- onset focal neurological symptoms together with the provisional diag-

nosis of stroke by the attending physician (Table 1).

The condition of the patient was quantified by calculating the pre- dictive scores with initial vital signs at triage station and the vital signs after ED resuscitation and treatment just before ambulance departure

Table 1

Definition of en-route deterioration Definitions

Physiological deterioration

Respiratory arrest or cardiac arrest Desaturation (SpO2 drop >= 5% or SpO2 b 90%) Systolic hypotension

Cardiac arrhythmia Bradyarrhythmia (b50/min) Tachyarrhythmia (N 140/min)

Neurological deterioration (drop in GCS >= 3 except intubated)

Hypothermia (b35?C)

(Table 2). The data handling and score calculation were performed by an independent data analyst who was blinded to patients’ outcome. A modified SCS was adopted in the study, with exclusion of the item “Prior to current illness, spent some part of daytime in bed” because it appeared irrelevant to the prediction of en-route complications. En- route clinical status was recorded by escort personnel on a specially de- signed IFT form in a prospective manner.

Definition of en-route deterioration

En-route deterioration was defined according to the previous study by Kanter et al, as follows: (1) respiratory arrest or cardiac arrest,

(2) desaturation (SpO2 drop >= 5% or SpO2 b 90%), (3) systolic hypoten- sion, (4). cardiac arrhythmia (pulse b 50/min or N 140/min), (5) neuro- logical deterioration (drop in GCS >= 3 except intubated), and

(6) hypothermia (b 35?C) [2,3,4] (Table 1).

Statistical analysis

The statistical software adopted was IBM SPSS Statistics for Win- dows, Version 22.0 (IBM Corp, Armonk, NY). Continuous data were expressed as means and standard deviation with comparisons per- formed by independent-sample t test. Categorical variables were expressed as frequencies and percentages. Parameters were compared between the group with en-route complications and without complica- tions by Fisher exact test or ?² test where appropriate. Boxplots were created to illustrate the difference in 3 scores in complicated vs uncom- plicated cases and for the scores calculated from vital signs in triage vs at ambulance departure.

The comparison of accuracy of predictive scores was assessed by re- ceiver operating characteristics (ROC) curves. The true-positive rate was plotted against the false-positive rate with nonparametric method, and the area under the curve (AUC) was calculated along with its 95% confidence interval (CI). By comparing the AUC of different scoring sys- tems based on vital signs at triage and at departure, discriminatory ca- pacities of these scores on predicting en-route complications could be evaluated. The level of significance was set at 5%.

The study was exempted from ethical approval because it was a ret-

rospective study with no impact on patient management.

Results

During the study period, 659 cases were included. The mean age was

59.8 years. Ninety-one (13.8%) cases were physician led, whereas 568 (86.2%) cases were nurse led. Endotracheal intubation was performed in 57 (8.6%) cases. Among the 659 cases, 17 (2.6%) cases had en-route complications. The complication rate in physician-led transport (2.2%) was similar to that in nurse-led transport (2.6%). All intubated cases had no en-route complications during transport. Physiological charac- teristics at triage and at departure showed no significant difference be- tween cases with en-route complications and uncomplicated cases, except that the mean diastolic blood pressure at triage was significantly different (complicated case, 56 +- 5 mm Hg; uncomplicated case, 80 +-

Table 2

Components of MEWS, HOTEL and SCS

2.1 MEWS [7-9]

Original design aims: to determine degree of illness by physiological variables

	Points	Systolic blood pressure (mm Hg)	Heart rate (beats per minute)	Respiratory rate (breaths per minute)	Temperature (?C)	Level of consciousness (AVPU)
	3	b 70
	2	70-80	b40	b9	b35
	1	81-100	41-50		35.1-36
	0	101-199	51-100	9-14	36.1-38	Alert
	1		101-110	15-20	38.1-38.5	React to voice
	2	>= 200	111-130	21-29	N 38.5	React to pain
	3		N 130	>=30		Unresponsive
	A for alert, V for verbal, P for pain, U for unresponsive
	2.2 HOTEL score [11]
	Original design aims: to predict early mortality between 15 min and 24 h after admission
	Parameter				Points
	Systolic blood pressure b100 mm Hg				1
	Oxygen saturation b90%				1
	Temperature b 35?C				1
	Abnormal ECG result (by computerized ECG interpretation program)				1
	Unstable to stand unaided				1
	2.3 SCS [10]
	Original design aims: to predict 30-d mortality after admission
	Parameter				Points
	Age (y)
	Men b50 or women b55				0
	Men 50-75 or women 55-75				2
	N 75				4
	Systolic blood pressure (mm Hg)
	N 100				0
	N 80-100				2
	N 70-80				3
	<= 70				4
	Pulse rate N systolic blood pressure				2
	Temperature b 35?C or >= 39?C				2
	Respiratory rate (/min)
	<= 20				0
	21-30				1
	N 30				2
	Oxygen saturation (%)
	>= 95				0
	90-94				1
	b 90				2
	Breathless on presentation				1
	Abnormal ECG results (by computerized ECG interpretation program, does not include sinus bradycardia and tachycardia)				2
	Diabetes (type I or II)				1
	Coma without intoxication or overdose				4
	Altered mental status without coma, intoxication or overdose, and aged >= 50 y				2
	New stroke on presentation				3
	Unable to stand unaided, or a nursing home resident				2
	Prior to current illness, spent some part of daytime in beda				2
	a Not included in the calculation of SCS in the study.

24 mm Hg; P = .001). The status of nursing home residence of patients was also significantly different (complicated case, 19%; uncomplicated case, 41.2%; P = .032) (Table 3).

Fig. 1 illustrated the scores of uncomplicated vs complicated trans- port. Complicated transport had generally higher scores which ap- peared more pronounced in MEWS and SCS calculated by vital signs at triage. Scores calculated by vital signs at triage were higher compared with those at ambulance departure.

The ROC curves of various scores in predicting en-route complica- tions were shown in Fig. 2. The AUC of MEWS at triage was 0.662 (95% CI, 0.547-0.777), and that at departure was 0.479 (95% CI, 0.34- 0.617). The accuracy of MEWS at triage was better than that at depar- ture (P = .049). The AUC of HOTEL at triage was 0.613 (95% CI, 0.457- 0.769), and that at departure was 0.597 (95% CI, 0.447-0.747). The AUC of SCS at triage was 0.6 (95% CI, 0.49-0.709), and that at departure

was 0.568 (95% CI, 0.455-0.68). The comparison of AUCs of HOTEL and SCS at triage and at departure showed no significant difference (Table 4).

Discussion

Various Risk stratification scores had been developed and validated for emergency patients in the setting of ED, in-patient wards, and criti- cal care units. Most of the predicted outcome variables were in-patient clinical deterioration or mortality. A recent study demonstrated that SCS and HOTEL score could predict mortality after admission with accept- able precision and excellent discrimination [12]. MEWS purely depends on physiological factors which are easy and convenient to assess, and it is commonly adopted in many in-patient wards for continuous moni- toring of patients to identify change of clinical status and for early

Table 3

Characteristics of the cohort with interhospital transport

Parameters	All (N = 659)	Transport with en-route complication (n = 17)	Uncomplicated transport (n = 642)	P value
Age, y (mean, SD)	59.8, 19.9	70.4, 13.8	59.5, 20.1	.120
Sex, male	376 (57.1%)	13 (76.5%)	363 (56.5%)	.136
Escort personnel				1.000
Physicians	91 (13.8%)	2 (11.8%)	89 (13.9%)
Nurses	568 (86.2%)	15 (88.2%)	553 (86.1%)
Traumatic	167 (25.5%)	5 (29.4%)	162 (25.4%)	.778
Intoxication	14 (2.1%)	0 (0%)	14 (2.2%)	1.000
Endotracheal intubated	57 (8.6%)	0 (0%)	57 (8.9%)	.386
Triage vital signs
Systolic BP (mean, SD)	135, 40	97, 20	136, 39	.116
Systolic BP b90 mm Hg	188 (13.4%)	5 (29.4%)	83 (12.9%)	.063
Diastolic BP (mean, SD)	80, 24	56, 5	80, 24	.001
Diastolic BP b60 mm Hg	135 (20.5%)	6 (35.3%)	129 (20.1%)	.132
Heart rate (mean, SD)	92, 22	103, 28	82, 21	.426
Heart rate >= 120/min	95 (14.4%)	2 (11.8%)	93 (14.5%)	1.000
Body temperature, ?C (mean, SD)	36.5, 0.8	35.6, 1.2	36.5, 0.8	.187
Body temperature >= 38?C or <= 35?C	49 (7.4%)	1 (5.9%)	48 (7.5%)	1.000
Respiratory rate >= 25/min	75 (11.4%)	2 (11.8%)	73 (11.4%)	1.000
Oxygen saturation SpO2 b 94%	74 (11.2%)	4 (23.5%)	70 (10.9%)	.112
GCS b 15	237 (36%)	6 (35.3%)	231 (36%)	1.000
AVPU level of consciousness				.709
Alert	474 (72.5%)	14 (82.4%)	460 (72.2%)
Verbal	37 (5.7%)	1 (5.9%)	36 (5.7%)
Pain	106 (16.2%)	1 (5.9%)	105 (10.5%)
Unconscious	37 (5.7%)	1 (5.9%)	36 (5.7%)
Departure vital signs
Systolic BP (mean, SD)	137, 35	123, 51	138, 35	.070
Systolic BP b90 mm Hg	61 (9.3%)	4 (23.5%)	57 (8.9%)	.063
Diastolic BP (mean, SD)	80, 21	86, 32	80, 20	.726
Diastolic BP b60 mm Hg	136 (20.6%)	5 (29.4%)	131 (20.4%)	.365
Heart rate (mean, SD)	84, 19	81, 21	87, 19	.409
Heart rate >= 120/min	73 (11.1%)	2 (11.8%)	71 (11.6%)	1.000
Body temperature, ?C (mean, SD)	36.5, 0.7	36.3, 0.6	36.5, 0.8	.522
Body temperature >= 38?C or <= 35 ?C	498 (75.6%)	12 (70.6%)	486 (75.7%)	.578
Respiratory rate >= 25/min	114 (17.3%)	2 (11.8%)	112 (17.4%)	.750
Oxygen saturation SpO2 b 94%	39 (5.9%)	2 (11.8%)	37 (5.8%)	.266
GCS b 15	276 (41.9%)	7 (41.2%)	269 (41.9%)	1.000
AVPU level of consciousness				.571
Alert	453 (70.6%)	13 (81.3%)	440 (70.3%)
Verbal	42 (6.5%)	1 (6.2%)	41 (6.5%)
Pain	79 (12.3%)	2 (12.5%)	77 (12.3%)
Unconscious	68 (10.6%)	0 (0%)	68 (10.9%)
Abnormal ECG result	370 (56.1%)	11 (64.7%)	359 (55.9%)	.662
Ambulatory status				.980
Trolley	57 (8.7%)	1 (5.9%)	56 (8.8%)
Wheelchair	37 (5.6%)	1 (5.9%)	36 (5.6%)
Ambulatory	561 (85.6%)	15 (88.2%)	546 (85.6%)
Nursing home resident	128 (19.5%)	7 (41.2%)	121 (19%)	.032
Diabetes mellitus	136 (20.9%)	4 (23.5%)	132 (20.8%)	.764
Breathless on presentation	83 (12.7%)	3 (17.6%)	80 (12.6%)	.466
New-onset stroke	244 (37.1%)	6 (35.3%)	238 (37.1%)	1.000

detection of clinical deterioration. However, none of the existing scores are dedicated or validated to predict en-route complications in IFT. Pre- diction and anticipation of en-route deterioration of patients of IFT would be essential in patient preparation, choice of escort personnel, and resource allocation. It would be more objective to achieve a clinical prediction score to predict the risk of transport rather than the clinical judgment of the attending physician alone.

Nevertheless, our study demonstrated that none of the scores had achieved satisfactory accuracy in predicting en-route complications for patients received IFT. Both scores based on physiological variables (MEWS and HOTEL score) and scores with additional variables on the present illness and premorbid status (SCS) showed no difference in accu- racy to predict en-route complications. This was consistent with previous studies [2,6]. Lee et al [2] had adopted MEWS and TISS-28 in prediction of en-route complications in IFT patients. The AUC of MEWS was demon- strated to be 0.71, which was comparable to our study (0.66). TISS-28 was nearly nondiscriminatory with AUC of 0.53. The fact that most existing scores did not accurately predict en-route complications is not

difficult to understand. The derivation of the score was based on multivar- iate models to predict a distinct outcome, either in-patient deterioration or mortality, but not en-route complications. Furthermore, the cohort is not comparable; the scores are valid in the setting of general in-patient wards or EDs but not limited to the patients for interhospital transport.

It appears that adoption of existing risk stratification scores would not achieve adequate accuracy to discriminate the complicated patients for IFT. A clinical prediction score derived with primary outcome to pre- dict deterioration during transport would be required. Looking into in- dividual physiological variables and components of the scores from Table 3, there were no strong predictive variables identified. Unless we could identify other highly predictive variables or otherwise, deriva- tion of an accurate model would be considered impractical. Possible fac- tors to be explored include patients’ preexisting comorbidities, the nature and severity of the present illness. and the resuscitation or inter- ventions performed in ED before transfer.

Among our evaluated scores, the best-performingd one was MEWS using vital signs at triage, which yielded an AUC of 0.66. In general,

Fig. 1. Boxplot of various scores in uncomplicated vs complicated transport.

scores calculated from vital signs at triage would perform better than those from ambulance departure. This is likely due to effective resusci- tation in the ED before transfer. This finding echoed the hypothesis that resuscitative treatment in ED may be one of the predictive variables for en-route complications. In addition, one may think that the worst vital signs in ED may be even more predictive; however, we could not evaluate this because of the limitation of the data set.

Fig. 2. receiver operating characteristics curves of predictive scores.

Table 4

Comparison of accuracy of predictive scores by ROC curves

Scores	AUC (95% CI)	P valuea	Comparison of triage vs departure
MEWS at triage	0.662 (0.547-0.777)	.027	.049
MEWS at departure	0.479 (0.34-0.617)	.773
HOTEL at triage	0.613 (0.457-0.769)	.121	NS
HOTEL at departure	0.597 (0.447-0.747)	.185
SCS at triage	0.6 (0.49-0.709)	.172	NS
SCS at departure	0.568 (0.455-0.68)	.356

^a P value calculated for null hypothesis of nondiagnostic (AUC = 0.5). NS = nonsignificant.

It was found that, for the cases with endotracheal intubation, in con- trast with conventional concept, the risk of en-route complication was low. The complication rate in physician-escorted cases was similar to that in nurse-escorted cases, although physician-escorted cases were expected to be more complicated and unstable. This reflected that suit- able personnel with adequate training and experience could lower the risk of en-route complication.

The overall complication rate in this study was 2.6%, which was lower than that in the studies by Lee et al [2] and Singh et al [5]. One of the rea- sons would be the lower proportion of critically ill patients being escorted to other hospitals, which was caused by the availability of intensive and cardiac care unit in the local hospital. In addition, we had not evaluated the compliance to the local IFT guideline and appropriateness of physician- or nurse-led transport, and therefore, overescort by health care professional may contribute to the apparently low complication rate.

Limitation

The study was based on retrospective data of a single center. There could be information bias of missing data in the IFT form. The chronic health status, premorbid status, and ambulatory status were retrieved from electronic records, which could be incomplete or missing. Escorts led by paramedics were not included in this study, whereas the compli- ance to local guideline on whenever the patients should be escorted by physicians, nurses, or paramedics was questionable. In terms of generaliz- ability, the cohort of patients for IFT was highly variable among hospitals, both locally and internationally. The cohort and complication rate would vary greatly depending on the availability of sophisticated care in the local hospital such as cardiac care and intensive care. The accuracies of the scores may not be validated in another local hospital of different setting.

Conclusion

None of the scores had good accuracy in prediction of en-route com- plications during IFT. MEWS at triage was among the best one already but was not ideal.

References

1. Feazel L, Schlichting AB, Bell GR, Shane DM, Ahmed A, Faine B, et al. Achieving re- gionalization through rural interhospital transfer. Am J Emerg Med 2015;33(9): 1288-96.
2. Lee LLY, Yeung KL, Lo WYL, Lau YSC, Tang SYH, Chan JTS. Evaluation of a Simplified Therapeutic Intervention Scoring System (TISS-28) and the Modified Early Warning Score (MEWS) in predicting physiological deterioration during inter-facility trans- port. Resuscitation 2008;76(1):47-51.
3. Kanter RK, Tompkin JM. Adverse events during interhospital transport: physiological de- terioration associated with pretransport severity of illness. Pediatrics 1989;84(1):43-8.
4. Britto J, Nadel S, Maconochie I, Levin M, Habibi P. Morbidity and severity of illness during interhospital transfer: specialized paediatric retrieval team. BMJ 1995; 311(7009):836-9.
5. Singh JM, MacDonald RD, Ahghari M. critical events during land-based interfacility transport. Ann Emerg Med 2014;64(1):9-15.
6. Brabrand M, Folkestad L, Clausen NG, Knudsen T, Hallas J. Risk scoring systems for adults admitted to the emergency department: a systematic review. Scand J Trauma Resusc Emerg Med 2010;18:8.
7. Subbe CP, Kruger M, Rutherford R, Gemmel L. Validation of a Modified Early Warn- ing Score in medical admissions. Q J Med 2001;94(10):521-6.
8. Olsson T, Terent A, Ling L. Rapid emergency medicine score: a new prognostic tool for in-hospital mortality in nonsurgical emergency department patients. J Intern Med 2004;255(5):579-87.
9. Lam TS, Mak PSK, Siu WS, Lam MY, Cheung TF, Rainer TH. Validation of a Modified Early Warning Score (MEWS) in emergency department observation ward patients. Hong Kong J Emerg Med 2006;13(1):24-30.
10. Kellett J, Deane B. The Simple Clinical Score predicts mortality for 30 days after ad- mission to an acute medical unit. Q J Med 2006;99(11):771-81.
11. Kellett J, Deane B, Gleeson M. Derivation and validation of a score based on hypotension, oxygen saturation, low temperature, ECG changes and loss of independence (HOTEL) that predicts early mortality between 15 min and 24 h after admission to an acute medical unit. Resuscitation 2008; 78(1):52-8.
12. Straede M, Brabrand M. External validation of the Simple Clinical Score and the HOTEL score, two scores for predicting short-term mortality after admission to an acute medical unit. PLoS One 2014;9(8), e105695.