You are on page 1of 12

Optimization of a Triage Model for Outbreaks of Ebola

Elizabeth Halter
2014-2015

1|Page

Table of Contents
Abstract..Page 3
Introduction Page 3
Problem...... Page 3
Hypothesis.. Page 3
Methods..Page 3- Page 7
Service Minutes..Page 7
Measuring Wait TimesPage 7- Page 8
Fatality Function.Page 8
Harvesting the Data.Page 9- Page 10
Analysis of the Data.Page 11
Conclusions..Page 11
ReferencesPage 12

2|Page

Abstract
In this project, emergency room triage procedures were analyzed, and statistical models of Ebola patients
were considered. Patient service data was taken from a typical, overwhelmed emergency room
(Providence Hospital, Washington DC) and used to create a statistical model of triage procedures and
their impact on waiting room contact time. Ebola infection was modeled to analyze impact of exposure. A
function was derived to determine overall fatalities in an emergency room. The function was optimized at
various Ebola outbreak levels to minimize total fatalities. The results are somewhat counterintuitive, as
data show that even when 5% of the sick population has Ebola, those patients should still be treated as
flu-like to minimize fatalities. The results offer emergency rooms a tool to combat the fear of the spread
of Ebola.
Introduction
Ebola Hemorrhagic Fever is a growing problem in our society. As distant countries become increasingly
interconnected, exposure to diseases that were once rare and relatively contained are now being carried to
new pools of possible victims. Last year, Ebola was carried to a new and very unexpected place by plane:
Dallas, Texas. All of a sudden, Ebola was a looming and very real threat to the American people. With
the first patient came fear and uncertainty. The threat of an Ebola outbreak was, and is, a very real one.
The question on everyones mind has become: What do we do if it starts to spread? This project seeks
to answer a critical part of that question.

Problem
Instead of a fear-based response to Ebola in a healthcare setting, a modification is proposed to the existing
Emergency Room Triage System based on statistical analysis of standard healthcare procedure.

Hypothesis
If current triage procedure is modified to allow for flexibility of rating of potential Ebola patients, then
fatalities can be minimized by prioritizing patients based on the percentage of the population that has
Ebola.
Methods
The triage procedure most widely used in the US is the ESI levelbased triage, which requires giving patients an ESI level from 1 to
5 based on an initial, front-desk assessment. ESI 1 patients are the
most critical and ESI 5 patients are the least. The decision-making
tree for ESI levels is to the right.
Triage is widely subjective. It is a complex game of trying to
predict patient need based on very little and often inconclusive data
received upon a patients arrival. ERs often assign their most
experienced and best nurse to triage, as the process is risky and
unclear. The nurse has to make predictions and decisions as to who
needs to be seen, and who can wait. These decisions are very
3|Page

difficult to make based on an initial and brief assessment. The actual triage process is very unclear and the
lines between the ESI levels are blurry, with the exception being ESI 1, which is assigned to a patient who
is dying upon triage assessment. Assigning a patient to ESI 2 is risky, as the nurse is taking a bet as to
whether or not the patient will rapidly deteriorate. 2s are high risk, meaning that the patient probably
shouldnt wait, but might be able to, allowing patients who are dying to be seen first.
Out of all patients seen (US national averages):
-

Around 2% are given ESI 1


Around 25% are given ESI 2
Around 35% are given ESI 3
Around 28% are given ESI 4
o 70% of ESI 4s exhibit flu-like symptoms
Around 10% are given ESI 5

In a typical ER, there are a limited number of bedded rooms that patients can be treated in. So, if an ESI 4
is being seen in a room and an ESI 1, 2, or 3 comes into the waiting room and there are no available beds,
the ESI 4 will be put on a cot in the hall so that the patient with a higher ESI can be treated first. ESI 2s
are discharged from rooms to allow ESI 1s to be seen. This is why wait times are dependent on a patients
ESI, as ESI level is the ultimate determinate of a patients path to care in the ER. ESI 4s are very low
priority, and must therefore wait until all patients with higher ESIs in the ER have been seen.
Wait times vary from ER to ER, but the general trend is the same; ESI 1s do not wait; ESI 2s wait for 10
to 20 minutes; ESI 3s wait up to four hours; and ESI 4s can wait up to 24 hours.
To build the model of an ER, a search was made for different sizes and wait times of ERs around the
country. After looking at many different ERs, Providence Hospital in Washington DC was selected, as it
is an overcrowded hospital that deals with many, many patients. Its average wait times per patient before
being placed in a bed are:
-

ESI 1: 0 minutes
ESI 2: 60 minutes (1 hour)
ESI 3: 180 minutes (3 hours)
ESI 4: 520 minutes (8 hours and 40 minutes)
ESI 5: Up to 12 hours, or leave without being seen

It takes about one hour for a patient in a bed to be seen and discharged (either admitted into the hospital,
sent into surgery, or sent home).The ER at Providence Hospital has 32 beds. This means that in a twelvehour time period, the hospital will see a total of 384 patients:
60
32
1

12

60

Of those 384 patients:


-

7.68 will be ESI 1


96 will be ESI 2

4|Page

= 384 patients

134.4 will be ESI 3


107.52 will be ESI 4
38.4 will be ESI 5
Knowing this, a relationship can
be modeled between ESI level and
average total wait time. This
relationship is shown with ESI
level as the independent variable
and total wait (average wait to be
seen + 60 bed minutes) as the
dependant, allowing total wait
time to be determined by assigned
ESI.

ESI vs. Total Time


500

y = 26.549e0.7267x

400
300
200

At this point, we can start to


attempt to show the effects of
Ebola on a waiting room.

100
0

Ebola, in its early stages of


development in the human body,
ESI of Patient
manifests very similarly to
influenza. Both diseases cause
high fever, muscle aches, fatigue, nausea, and vomiting. There is no quick (<1 day) test to differentiate
between Ebola and flu patients. The ESI triage procedure is not set up to distinguish between flu patients
and Ebola patients, as it has never needed to do so. Therefore, if an early-stage Ebola patient comes into
an ER with a flu patient, both will be labeled ESI 4 and will sit in the waiting room for an extended
amount of time.
1

1.5

2.5

3.5

In fact, Ebola and Influenza have similar R0 (basic


reproduction) values, meaning that both spread at
about the same rate. Both have an R0 value of
around 2, meaning that over the course of the
disease, the host will infect an average of two other
people.
Using this R0 value and the equation for effective
contact rate (B=Y x P), a relationship between an
Ebola patients total wait time and number of people
infected and possibly killed can be shown.
In the Effective Contact Rate Equation, B = Y x P:
-

B = effective contact rate (total people


infected)
Y = total contact (total wait time)

Ebola Fatalities vs.


Wait Times
3

Number of Fatalities

Patient Wait Time (minutes)

600

y = 0.8813ln(x) - 2.5974

2.5

1.5

1
60

5|Page

260

460

Patient Wait Time (minutes)

ESI 1 Likely Fatalities vs.


Wait Times

With this equation, a correlation between the total


wait time for an Ebola patient and the number of
possible fatalities their visit might cause can be
drawn.

y = -0.001x2 + 0.178x - 7.075

70%

- If the patient waits for no time, their visit


likely causes one fatality: themselves.
- If the patient waits as long as an ESI 3
typically waits, then their visit likely causes two
fatalities: themselves and one other that they infect.
- If the patient waits as long as an ESI 4
waits, then their visit likely causes three fatalities:
themselves and two others that they infect.

60%
50%
40%
30%
20%
10%
0%
60

70

80

90

This information can be used to create a function


that models likely kills per Ebola patient versus
average total wait time.

Patient Wait Time (minutes)

However, people other than Ebola patients die in ER waiting rooms. In fact, around 3% of people treated
in emergency rooms die.
Since ER deaths are time-sensitive, correlations between total wait time and deaths for ESI 1s and ESI 2s
need to be drawn if the model is to be accurate.
ESI 1 patients are described as patients who are dying and need immediate life-saving treatment.
Therefore, the relationship between total wait times and percentage of ESI 1s who die is logarithmic in
nature. When an ESI 1 patient does not wait at all, meaning that their total wait is 60 minutes, they do not
die. They are out of the ER in enough
time to get into surgery or get admitted
into the hospital. If they do die, they die
in other areas. However, as total wait
time reaches 70 minutes (10 minutes in
60%
the waiting room) around 50% of ESI 1s
50%
y = 0.001x - 0.0577
will die. At some point, however, almost
all of the patients are fatalities. A small
40%
percentage of ESI 1s will not die at an
30%
hour and a half total wait. This
relationship is represented by the function
20%
to the right.

ESI 2 Likely Fatalities vs. Wait


Times

ESI 2 patients are described as patients


who are risky and may develop into fatal
cases, but are not dying upon entry. At
the national average wait time, only 1-2%
6|Page

Percentage of Patients Killed

Percentage of Patient Fatalities

90%
80%

P = transmission risk (R0 value)

10%
0%
60

160

260

360

460

Patient Wait Time (minutes)

560

of ESI 2s die. Using this data, we can generate a relationship between average total wait time and
percentage of ESI 2 patients who die.
ESI 3s in general do not die in the ER, because
if the triage nurse had determined that they had
any potential to die, they would have been given
a two.

Wait Minutes per Patient


At Providence Hospital
ESI Level (# of Patients)
ESI 1 (7.68)

Total Wait
Minutes
(Wait x # of
Patients)
0

Service Minutes
ESI 2 (96)
60
5760
The ER processes a certain, set number of
ESI 3 (134.4)
120
16128
patients in a twelve hour period. The assumption
ESI 4 (107.52)
520
55910.4
is that the ER is processing at its maximum
capacity, 384 patients every 12 hours. It is also
Flu-like Patients(38.4)
520
39137.28
assumed that the total number of people coming
Grand Total
77806.08
Wait Minutes
into the ER is the same, no matter what ESI
levels are assigned. Based on these two
assumptions, we can calculate the total number of wait minutes accumulated by all patients in a twelvehour period, which must remain constant.
= 77806 = 1 1 + 2 2 + 3 3 + 4 4 +
where is total wait time over a 12-hour period, and therefore the number of service minutes.
ESI 5s are not included, as flu-like patients are currently given ESI 4, and the proposal does not suggest
downgrading their ESI level, making ESI 5 waits constant and therefore irrelevant.
The total number of service minutes available to ESI 4 and above is 77806 service minutes. No matter
what ESI I assign to an Ebola patient, the total wait minutes for all ESI levels above ESI 4 must remain
constant.
Measuring Wait Times
As wait time is a function of ESI level, wait times can be derived from the equation:
= 26.54 0.726(+)
Where:
-

is total wait time for that ESI level


ESI is ESI level
C is the constant of proportionality that is used to shift the ESI to the right on the ESI vs. Wait
Times Graph

Total service minutes remain constant, so to find the average total wait time for a certain ESI level, the
patients ESI level added to a constant of proportionality is put into the Wait vs. ESI equation.
Essentially, the constant of proportionality (C) represents a slide on the Wait vs. ESI graph that must be
found through iteration. The equation to calculate C at each assigned flu-like ESI is:

77806 = 1 26.54 0.726(1+) + 2 26.54 0.726(2+) + 3 26.54 0.726(3+) + 4 26.54


0.726(4+) + 26.54 0.726(+)

7|Page

Which is derived from the total service minutes equation.


When the known PEsi are plugged in:
77806 = 203.8 0.726(1+) + 2548 0.726(2+) + 3567 0.726(3+) + 856.1 0.726(4+)
+ 1997.5 0.726(+)
To find C for each ESI that is assigned to flu-like patients (and therefore Ebola patients as well), the
assigned ESI is plugged in to the ESIEbola and C is found through goal-seeking iteration. Once C is known
for a certain assigned ESI level, the average total wait times can be calculated using the Wait Times
equation and be used in the Fatality Function.

Fatality Function
The number of fatalities in the ER depends on the average total wait times per ESI level patient, as stated
earlier. The total fatalities can be expressed by:
= 1 + 2 +
where the K represents the number of kills caused by each group.
- K1 is defined as total fatalities within the ESI 1 patients.
- K 2 is defined as total fatalities within the ESI 2 patients.
- KEbola is defined as total fatalities within the Ebola patients.
The hypothesis is that KT can be minimized by varying the ESI level of flu-like patients.
Kills can be expressed for each group with the kill functions generated earlier.
Therefore K1 can be expressed as:
1 = (0.001 1 2 + 0.178 1 7.075) 1
Where P1 is the number of non-flu-like ESI 1 patients, a known number.
When the WEsi is plugged in, the complete ESI 1 fatalities as a function of C is:
1 = 0.001 (26.54 0.726(1+) )2 + 0.178 26.54 0.726(1+) 7.075
K2 as a function of C can be expressed as:
2 = (0.001 26.54 0.726(2+) 0.057) 2
with P2 being the total number of non-flu-like ESI 2 patients.
KEbola can be expressed as a function of C as:
= (0.8813 (26.54 0.726(+) ))
with P2 being the total number of Ebola patients.
Therefore, the total number of fatalities at an assigned flu-like ESI level is expressed as the sum of the
three:
= (0.001 (26.54 0.726(1+) )2 + 0.178 26.54 0.726(1+) 7.075) 1
+ (0.001 26.54 0.726(2+) 0.057) 2
+ (0.8813 (26.54 0.726(+) ))

8|Page

Harvesting the Data


Obviously, different ESI levels are more appropriate during different levels of outbreak. If there is only a
.01% chance that a flu-like patient has Ebola, it makes no sense to treat all flu-like patients as ESI .5 and
treat them before gunshot patients. Conversely, it makes no sense to assign flu-like patients an ESI of 4
and allow them to sit for hours when all of the flu-like patients have Ebola (100% likelihood). To test the
hypothesis, it is necessary to determine the best modification at differing levels of outbreak to allow ERs
to respond appropriately to all scenarios.
In order to determine what ESI level is most appropriate to assign to flu-like patients to minimize likey
fatalities, two figures need to be varied:
- ESI level assigned to flu-like patients
- Percentage likelihood that a flu-like patient has Ebola
ESI level is varied in the C iterations, where C is dependent on the assigned flu-like ESI.
To allow flexibility in the ESI level system, I propose that intermediate ESI levels be added to help better
sort out and prioritize flu-like patients.
ESI 0.5
At this ESI level, flu-like patients are seen before patients with gunshot wounds.
ESI 1
At this ESI level, flu-like patients are treated as equivalent to patients with gunshot wounds.
ESI 1.5
At this ESI level, flu-like patients are treated after patients with gunshot wounds, but before a
patient that may be having a heart-attack.
ESI 2
At this ESI level, flu-like patients are treated as equivalent to patients who may be having a heart
attack.
ESI 2.5
At this ESI level, flu-like patients are treated after patients who may be having a heart-attack, but
before a patient who has a broken leg.
ESI 3
At this ESI level, flu-like patients are treated as equivalent to patients who have broken bones.
ESI 3.5
At this ESI level, flu-like patients are treated after patients with broken bones, but before patients
in need of stitches.
ESI 4
At this ESI level, flu-like patients are treated as equals to patients in need of stitches.
To vary the percentage likelihood that a flu-like patients have Ebola, I use the equation:
= # %
within the KT equation.
This allows the generation of the final data field:

9|Page

% Likelyhood a
Flu-Like
Patient has
Ebola
ESI Assigned to Flu-Like Patients
0.50%
1.00%
1.50%
2.00%
2.50%
3.00%
3.50%
4.00%
4.50%
5.00%
5.50%
6.00%
6.50%
7.00%
7.50%
8.00%
8.50%
9.00%
9.50%
10.00%
10.50%
11.00%
11.50%
12.00%
12.50%
13.00%
13.50%
14.00%
14.50%
15.00%
15.50%
16.00%
16.50%
17.00%
17.50%
18.00%
18.50%
19.00%
19.50%
20.00%
20.50%
21.00%
21.50%
22.00%
22.50%
23.00%
23.50%
24.00%
24.50%
25.00%
25.50%
26.00%
26.50%
27.00%
27.50%
28.00%
28.50%
29.00%
29.50%
30.00%

10 | P a g e

0.5
20.103
20.555
21.006
21.457
21.908
22.359
22.810
23.261
23.712
24.163
24.614
25.065
25.516
25.967
26.418
26.869
27.320
27.771
28.222
28.673
29.124
29.575
30.026
30.477
30.928
31.379
31.830
32.281
32.732
33.183
33.634
34.085
34.536
34.987
35.438
35.889
36.340
36.791
37.242
37.693
38.145
38.596
39.047
39.498
39.949
40.400
40.851
41.302
41.753
42.204
42.655
43.106
43.557
44.008
44.459
44.910
45.361
45.812
46.263
46.714

1
18.830
19.338
19.845
20.353
20.861
21.369
21.876
22.384
22.892
23.399
23.907
24.415
24.923
25.430
25.938
26.446
26.954
27.461
27.969
28.477
28.984
29.492
30.000
30.508
31.015
31.523
32.031
32.538
33.046
33.554
34.062
34.569
35.077
35.585
36.092
36.600
37.108
37.616
38.123
38.631
39.139
39.647
40.154
40.662
41.170
41.677
42.185
42.693
43.201
43.708
44.216
44.724
45.231
45.739
46.247
46.755
47.262
47.770
48.278
48.785

1.5
12.000
12.617
13.234
13.851
14.468
15.085
15.702
16.319
16.936
17.553
18.170
18.787
19.404
20.021
20.638
21.255
21.872
22.489
23.106
23.723
24.340
24.957
25.574
26.191
26.808
27.425
28.042
28.659
29.276
29.893
30.510
31.127
31.744
32.361
32.978
33.595
34.212
34.829
35.446
36.063
36.680
37.297
37.914
38.531
39.148
39.765
40.382
40.999
41.616
42.233
42.850
43.467
44.084
44.701
45.318
45.935
46.552
47.169
47.786
48.403

2
11.450
12.173
12.895
13.618
14.340
15.063
15.786
16.508
17.231
17.953
18.676
19.399
20.121
20.844
21.566
22.289
23.012
23.734
24.457
25.179
25.902
26.625
27.347
28.070
28.792
29.515
30.238
30.960
31.683
32.405
33.128
33.850
34.573
35.296
36.018
36.741
37.463
38.186
38.909
39.631
40.354
41.076
41.799
42.522
43.244
43.967
44.689
45.412
46.135
46.857
47.580
48.302
49.025
49.748
50.470
51.193
51.915
52.638
53.361
54.083

2.5
10.698
11.521
12.344
13.168
13.991
14.814
15.638
16.461
17.285
18.108
18.931
19.755
20.578
21.401
22.225
23.048
23.872
24.695
25.518
26.342
27.165
27.988
28.812
29.635
30.458
31.282
32.105
32.929
33.752
34.575
35.399
36.222
37.045
37.869
38.692
39.516
40.339
41.162
41.986
42.809
43.632
44.456
45.279
46.102
46.926
47.749
48.573
49.396
50.219
51.043
51.866
52.689
53.513
54.336
55.160
55.983
56.806
57.630
58.453
59.276

3
9.747
10.665
11.584
12.503
13.421
14.340
15.259
16.177
17.096
18.015
18.933
19.852
20.771
21.689
22.608
23.526
24.445
25.364
26.282
27.201
28.120
29.038
29.957
30.876
31.794
32.713
33.632
34.550
35.469
36.388
37.306
38.225
39.144
40.062
40.981
41.900
42.818
43.737
44.656
45.574
46.493
47.411
48.330
49.249
50.167
51.086
52.005
52.923
53.842
54.761
55.679
56.598
57.517
58.435
59.354
60.273
61.191
62.110
63.029
63.947

3.5
8.532
9.537
10.543
11.549
12.554
13.560
14.566
15.571
16.577
17.583
18.588
19.594
20.600
21.605
22.611
23.617
24.622
25.628
26.634
27.639
28.645
29.651
30.656
31.662
32.668
33.673
34.679
35.685
36.690
37.696
38.702
39.707
40.713
41.719
42.724
43.730
44.736
45.741
46.747
47.753
48.758
49.764
50.770
51.775
52.781
53.787
54.792
55.798
56.804
57.809
58.815
59.821
60.826
61.832
62.838
63.843
64.849
65.855
66.860
67.866

4
7.219
8.352
9.485
10.618
11.751
12.883
14.016
15.149
16.282
17.415
18.548
19.681
20.813
21.946
23.079
24.212
25.345
26.478
27.611
28.743
29.876
31.009
32.142
33.275
34.408
35.540
36.673
37.806
38.939
40.072
41.205
42.338
43.470
44.603
45.736
46.869
48.002
49.135
50.268
51.400
52.533
53.666
54.799
55.932
57.065
58.197
59.330
60.463
61.596
62.729
63.862
64.995
66.127
67.260
68.393
69.526
70.659
71.792
72.925
74.057

Analysis of the Data


Optimizing the assigned ESI level is fairly straightforward. The goal of the ER is to minimize fatalities.
Therefore, to determine which ESI should be assigned to flu-like patients at each % likelihood of Ebola,
we can look at which ESI causes the least number of deaths. The optimization of the data is represented
by the red line. The line represents the lowest likely death count at each % likelihood of Ebola. ERs
should assign the ESI shown on the chart that correlates to the levels of outbreak that their area is
experiencing.
Conclusion
Based on data analysis, the hypothesis
was confirmed. By allowing differing
ESI levels, including intermediate ESI
levels, to be assigned to flu-like
patients based on the percentage
likelihood that a flu-like patient has
Ebola, likely fatalities can be
minimized. This result is somewhat
counterintuitive, as the data show that
even when 5% of the sick population
has Ebola, flu-like patients should still
be treated as ESI 4 to minimize
fatalities. These data and models are
based on the current understanding of
the infection rates of Ebola as well as
current triage procedure. This same
strategy could be used to model other
outbreaks with modification only to
the Ebola Kill function. This system can be used by any hospital that uses the ESI triage system to help
respond to outbreak and minimize fear, uncertainty, and deaths in the Emergency Room.

11 | P a g e

References
"Academic Emergency MedicineVolume 8, Issue 1, Article First Published Online: 28 JUN
2008." Emergency Severity Index Triage Category Is Associated with Six-month Survival. N.p.,
n.d. Web. 26 Feb. 2015.
"Center for Studying Health System Change." HSC Research Brief No. 23. N.p., n.d. Web. 26 Feb.
2015.
"ER Wait Watcher." ProPublica. N.p., n.d. Web. 26 Feb. 2015.
"Healthcare 411." Emergency Severity Index (ESI) Implementation Handbook, 2012 Edition. N.p., n.d.
Web. 26 Feb. 2015.
Ramesh, Aruna C., and S. Kumar. "Triage, Monitoring, and Treatment of Mass Casualty Events
Involving Chemical, Biological, Radiological, or Nuclear Agents." Journal of Pharmacy and
Bioallied Sciences. Medknow Publications Pvt Ltd, n.d. Web. 26 Feb. 2015.
"Result Filters." National Center for Biotechnology Information. U.S. National Library of Medicine,
n.d. Web. 26 Feb. 2015.
"Seasonal Influenza Q&A." Centers for Disease Control and Prevention. Centers for Disease Control
and Prevention, 14 Aug. 2014. Web. 20 Feb. 2015.
"Simple Triage and Rapid Treatment (START)." Simple Triage and Rapid Treatment Course. N.p.,
n.d. Web. 26 Feb. 201

12 | P a g e

You might also like