Anaesthesia 2018 doi:10.1111/anae.

14319

Original Article
The organisation of critical care for burn patients in the UK:
epidemiology and comparison of mortality prediction models
A. P. Toft-Petersen,1,2 P. Ferrando-Vivas,3 D. A. Harrison,4 K. Dunn5 and K. M. Rowan6

1 Researcher, Departments of Clinical Medicine and Anaesthesia and Intensive Care, Aalborg University Hospital,
Aalborg, Denmark
2 Visiting Research Fellow, 3 Statistician, 4 Head Statistician, 6 Director, Intensive Care National Audit and Research
Centre (ICNARC), London, UK
5 Consultant, Adult Burn Service, University Hospital of South Manchester, Manchester, UK

Summary
In the UK, a network of specialist centres has been set up to provide critical care for burn patients. However, some burn
patients are admitted to general intensive care units. Little is known about the casemix of these patients and how it com-
pares with patients in specialist burn centres. It is not known whether burn-specific or generic risk prediction models
perform better when applied to patients managed in intensive care units. We examined admissions for burns in the Case
Mix Programme Database from April 2010 to March 2016. The casemix, activity and outcome in general and specialist
burn intensive care units were compared and the fit of two burn-specific risk prediction models (revised Baux and
Belgian Outcome in Burn Injury models) and one generic model (Intensive Care National Audit and Research Centre
model) were compared. Patients in burn intensive care units had more extensive injuries compared with patients in gen-
eral intensive care units (median (IQR [range]) burn surface area 16 (7–32 [0–98])% vs. 8 (1–18 [0–100])%, respectively)
but in-hospital mortality was similar (22.8% vs. 19.0%, respectively). The discrimination and calibration of the generic
Intensive Care National Audit and Research Centre model was superior to the revised Baux and Belgian Outcome in
Burn Injury burn-specific models for patients managed on both specialist burn and general intensive care units.
.................................................................................................................................................................
Correspondence to: A. P. Toft-Petersen
Email: anne.toft-petersen@icnarc.org
Accepted: 1 April 2018
Keywords: burns; critical care; epidemiology; hospital mortality; models: statistical; thermal injury

Introduction networks; Northern Ireland has a separate burn care

Annually, burn injuries are estimated to cause about
180,000 deaths worldwide and the associated morbidity
is considerable [1]. Each year, about 175,000 people
present to Emergency Departments in the UK with
burn injuries [2] and in 2010, 11,477 people were
admitted to acute hospitals in England with a primary
diagnosis of burn injury [3].
The framework for burn care in England and
Wales consists of four burn care operational delivery
organisation [4]. The system in England and Wales is
three-tiered with burn centres, burn units and burn
facilities [5]. Burn centres provide the highest level of
specialised burn care (including intensive care) and in
England and Wales there are 11 such centres which
admit adult patients; Northern Ireland has a single
burn centre which admits adults [4]. National Health
Service (NHS) guidelines specify that adult patients
should be referred to a burn centre if the burn surface

© 2018 The Association of Anaesthetists of Great Britain and Ireland 1

Anaesthesia 2018 Toft-Petersen et al. | Critical care organisation for burn patients in the UK
area is > 40% of total body surface area or with a burn
surface area > 25% of total body surface area with an
associated inhalation injury [6].
Universally agreed risk factors for mortality follow-
ing burn injury are age and burn surface area [7, 8]. In
addition, inhalation injury [9], full skin thickness
burns [10], burn mechanism [11], adjacent trauma [12],
comorbidities [13–15], drugs and alcohol [16], physio-
logical derangement [7, 17–19], sex [20–22] and obesity
[13, 23] have also been reported as risk factors for mor-
tality. Many burn-specific risk prediction models have
been developed; however, these have predominantly
been developed and validated in combined burn centre
and burn unit patient populations [24] and little is
known of their performance in burn patients admitted
to an intensive care unit (ICU).
Given the high level of recent interest in the care
of, and outcomes for, critically ill burn patients, the
objectives of this study were: to describe the character-
istics of patients with burn injuries admitted to ICUs
in England, Wales and Northern Ireland; to compare
those patients admitted to ICUs in burn centres vs.
non-burn centres; and to assess and compare the per-
formance of two burn-specific and one generic risk
prediction model in ICU patient populations.
Methods
The Case Mix Programme (CMP) is the national clini-
cal audit for adult intensive care in England, Wales
and Northern Ireland. The CMP database contains
pooled casemix and outcome data on consecutive
admissions to participating ICUs (both intensive care
and combined intensive care/high dependency units).
Casemix and outcome data are collected to precise
rules and definitions, by trained data collectors, and
undergo extensive local and central validation before
pooling. Details of the data collection and validation
have been reported previously [25]. The CMP database
has been independently assessed to be of high quality
[25] and support for the collection and use of patient-
identifiable data without consent has been obtained
under Section 251 of the NHS Act 2006.
Intensive care units participating in the CMP dur-
ing the period 1 April 2010 to 31 March 2016 were
included; ICUs located in hospitals with burn centre
status and which were the main location where burn
2 © 2018 The Association of Anaesthetists of Great Britain and Ireland
patients were provided intensive care were termed
burn ICUs, whereas ICUs located in hospitals without
burn centre status (and, therefore, not the usual loca-
tion for burn patients to receive intensive care) were
termed general ICUs.
Data for admissions with burn injury to eligible
ICUs with the following criteria were extracted: age
≥ 16 years; level 2 (high dependency) or level 3 (inten-
sive) care in the first 24 h after admission; and first
admission to ICU for this burn injury. For the CMP,
clinicians record the primary and secondary reasons
for admission using the Intensive Care National Audit
and Research Centre (ICNARC) coding method, a
hierarchical coding system specifically designed for
recording reasons for admission to intensive care [26].
For each identified patient, the following data
were extracted: burn surface area; presence of inhala-
tion injury; burn mechanism; source of admission;
urgency of surgery (emergent/urgent compared with
elective/scheduled); age; sex; assistance in daily activi-
ties; severe conditions in the past medical history;
acute severity of illness (assessed with the ICNARC
physiology score [27] and Acute Physiology And
Chronic Health Evaluation (APACHE-2 acute physiol-
ogy score [28]); organ support; sedation; highest level
of care during the first 24 h; treatment orders; length
of stay in ICU; length of stay in acute hospital; trans-
fer to other critical care unit; ICU mortality; and in-
hospital mortality.
Burn surface area was defined as the proportion of
the total body surface area that had second- or third-
degree burns. Inhalation injury was considered present
if coded as inhalation burns or inhalation injury and
mechanical ventilation was initiated in the first 24 h.
Burn mechanism was defined as either ‘heat’, ‘steam’,
‘electrical’ or ‘chemical’.
Duration of organ support was measured in calen-
dar days (00:00–23:59 or part thereof) for advanced res-
piratory, renal and dermatological support, as defined
by the Department of Health Critical Care Minimum
Data Set (CCMDS) [29]. Treatment orders were
recorded as active treatment withheld and/or with-
drawn. An admission was recorded as ‘transferred for
further critical care’ if the discharge destination was
either an ICU or high dependency unit. Intensive care
unit duration of stay was calculated as the number of
Toft-Petersen et al. | Critical care organisation for burn patients in the UK Anaesthesia 2018

days from the date and time of admission to discharge/
death. The total acute hospital duration of stay was cal-
culated as the number of days from the date of original
admission to an acute hospital to the date of ultimate
discharge from an acute hospital. Mortality was assessed
at discharge from ICU and at discharge from acute
hospital.
Casemix, activity and outcome for patients with
burn injury were described by type of ICU, either burn
or general. Differences in burn surface area between
burn and general ICUs were tested using the Wilcoxon
rank-sum test, and differences in the presence of
inhalation injury and mortality at ICU and acute hos-
pital discharge were tested with logistic regression.
Three risk predictions for acute hospital mortality
for each patient were calculated by applying the
revised Baux model [30], Belgian Outcome in Burn
Injury (BOBI) model [31], and the 2015 recalibration
of the ICNARC model [32] (see Appendix S1 for
details). The performance of the risk prediction models
was estimated by calculating their discrimination, cali-
bration and accuracy. The discrimination of the model
was assessed by the area under the receiver operating
characteristic curve (AUC). The calibration was
assessed both graphically (with predicted probability
on the x-axis and the observed outcomes on the y-axis
in 10 equal-sized risk groups) and by Cox’s calibration
regression [33] (linear recalibration of the predicted

812,510 admissions
to 222 ICUs

Admissions with
burns
(n = 1311)

Excluded (n = 170)
- Age < 16 years (n = 21)
- Re-admission/transfer (n = 127)
- No level 2/3 care within first 24 h (n = 22)

1141 first admissions

with burns to 155
ICUs

640 admissions to 501 admissions to

7 burn ICUs 148 general ICUs

Missing data (n = 8) Missing data (n = 30)

Analysed (n = 632) Analysed (n = 471)

Figure 1 Patient selection for inclusion.

© 2018 The Association of Anaesthetists of Great Britain and Ireland 3

Anaesthesia 2018 Toft-Petersen et al. | Critical care organisation for burn patients in the UK

Table 1 General characteristics of the study cohort. All values had less than 3% missing with the exception of pres-
ence of inhalation injury (7.2% missing). Values are mean (SD), number (proportion) or median (IQR [range]).

All ICUs Burn ICUs General ICUs

n = 1141 n = 640 n = 501
Age; years 48 (19.2) 47 (19.1) 49 (19.4)
Sex; male 827 (73%) 454 (71%) 373 (75%)
Burn surface area; % 10 (5–26 [0–100]) 16 (7–32 [0–98]) 8 (1–18 [0–100])
Inhalation injury present 502 (44%) 284 (44%) 218 (44%)
Burn mechanism
Chemical 92 (8.1%) 63 (9.8%) 29 (5.8%)
Electrical 57 (5.0%) 32 (5.0%) 25 (5.0%)
Heat 776 (68%) 492 (77%) 284 (57%)
Inhalation only 143 (13%) 16 (2.5%) 127 (25%)
Steam 73 (6.4%) 37 (5.8%) 36 (7.2%)
Source of admission
ED (same hospital) 673 (59%) 228 (36%) 445 (89%)
ED (other hospital) 301 (26%) 295 (46%) 6 (1.2%)
Ward (same hospital) 142 (12%) 93 (15%) 49 (9.8%)
Ward (other hospital) 25 (2.2%) 24 (3.8%) 1 (0.2%)
Surgery 152 (13%) 102 (16%) 50 (10%)
Emergent/urgent 132 (12%) 88 (14%) 44 (8.8%)
Elective/scheduled 20 (1.8%) 14 (2.2%) 6 (1.2%)
Assistance in daily activities
None 943 (83%) 529 (83%) 414 (83%)
Some 165 (15%) 100 (16%) 65 (13%)
Total 6 (0.5%) – 6 (1.2%)
Severe comorbid conditions
Cardiovasculara 7 (0.6%) 6 (0.9%) 1 (0.2%)
Respiratoryb 25 (2.2%) 12 (1.9%) 13 (2.6%)
Renalc 3 (0.3%) 2 (0.3%) 1 (0.2%)
Liverd 9 (0.8%) 7 (1.1%) 2 (0.4%)
Immunosuppression 19 (1.7%) 11 (1.7%) 8 (1.6%)
APACHE-2 acute physiology score 9 (6–13 [0–40]) 10 (7–13 [0–33]) 8 (5–12 [0–40])
ICNARC physiology score 16 (11–22 [0–60]) 17 (12–22 [0–54]) 15 (10–22 [0–60])
ICNARC acute hospital mortality probability; % 9 (3–25 [0–100]) 10 (4–25 [0–100]) 7 (3–26 [0–100])

ICU, intensive care unit; ED, emergency department; APACHE, Acute Physiology and Chronic Health Evaluation, ICNARC,
Intensive Care National Audit and Research Centre.
a
New York Heart Association functional classification, class 4.
b
Permanent shortness of breath with light activity due to pulmonary disease or home ventilation.
c
Chronic renal replacement therapy for irreversible renal disease.
d
Portal hypertension/biopsy proven cirrhosis/hepatic encephalopathy.
e
Steroid treatment/chemotherapy/radiotherapy/metastatic disease/lymphoma/acute or chronic myelogenous/lymphocytic leukae-
mia/multiple myeloma/congenital immunohumoral or cellular immune deficiency state/HIV/AIDS.

log odds). The calibration regression was expressed by
an intercept (intercept = 0 with perfect calibration)
and a slope (slope = 1 with perfect calibration). Accu-
racy was assessed by Brier’s score (the mean squared
error between outcomes and predictions). Performance
was compared overall and by type of ICU. We com-
pared the discrimination of the revised Baux and the
BOBI models to the ICNARC model as our objective
was to compare burn-specific and generic models.
Three sensitivity analyses were conducted to test
the robustness of the results to decisions in the
analysis. The first did not include patients solely with
electrical burns as burn surface area poorly captures
the actual internal damage resulting from electrical
burns [34] and this could contribute to poor fit. The
second did not include patients admitted between
April 2013 and March 2014 (the recalibration sample
for the ICNARC model) to avoid potential overfitting;
due to the small contribution of burn patients to the
overall ICNARC model recalibration sample, it was
not considered necessary to not include these patients
from the primary analysis. The third analysis recoded

4 © 2018 The Association of Anaesthetists of Great Britain and Ireland

Toft-Petersen et al. | Critical care organisation for burn patients in the UK Anaesthesia 2018

Table 2 Intensive care unit (ICU) support and outcomes for burn patients managed in burn and general intensive
care units. All values had less than 3% missing. Values are number (proportion) or median (IQR [range])

All ICUs Burn ICUs General ICUs

n = 1141 n = 640 n = 501
Dermatological supporta 585 (51%) 424 (66%) 161 (32%)
Duration of dermatological supportd; days 4 (2–13 [1–96]) 6 (2–19 [1–96]) 2 (1–4 [1–56])
Advanced respiratory supportb 875 (77%) 540 (84%) 335 (67%)
Duration of respiratory supportd; days 2 (1–9 [1–82]) 4 (2–13 [1–82]) 2 (1–3 [1–57])
Renal supportc 82 (7.2%) 58 (9.1%) 24 (4.8%)
Duration of renal supportd; days 4 (2–12 [1–41]) 7 (2–14 [1–41]) 3 (2–6 [1–18])
Sedated/paralysed during the first 24 h 848 (74%) 532 (83%) 316 (63%)
Level-3 care in the first 24 h 941 (83%) 576 (90%) 365 (73%)
Treatment restrictions 120 (11%) 78 (12%) 42 (8.4%)
ICU length of stay; days 2 (1–6 [0–97]) 3 (1–12 [0–97]) 1 (0–3 [0–65])
Acute hospital duration of stay; days 13 (4–34 [0–371]) 18 (6–40 [0–371]) 7 (2–24 [0–163])
Transferred to another ICU 163 (14%) 13 (2.0%) 150 (30%)
ICU mortality 184 (16%) 120 (19%) 64 (13%)
In-hospital mortality 241 (21%) 146 (23%) 95 (19%)

a
Days with major (e.g. > 30% body surface area affected) skin rashes, exfoliation or burns, or time receiving complex dressings
(skin dressings, open abdomen, vacuum dressings or multiple limb or limb and head trauma dressings).
b
Days receiving invasive mechanical ventilatory support (bilevel positive airway pressure applied via a translaryngeal tracheal tube
or tracheostomy; continuous positive airway pressure via a translaryngeal tracheal tube; or extracorporeal respiratory support).
c
Days receiving acute renal replacement therapy (e.g. haemodialysis, haemofiltration etc.) or renal replacement therapy for chronic
renal failure where other acute organ support is received.
d
Only for patients who had support.

patients initially admitted to a general ICU who were
subsequently transferred to another critical care unit
(assumed to be a burn ICU) within the same acute
hospital stay to explore the impact of using the initial
categorisation.
All analyses were performed using Stata/SE Ver-
sion 13 (StataCorp LP, College Station, TX, USA).
Results
Out of 812,510 patients admitted to 222 ICUs between
1 April 2010 and 31 March 2016, 1141 (0.14%) were
adult admissions for burn injury (Fig. 1). Patients with
burns were admitted to both burn and general ICUs
but burn ICUs admitted a mean (SD) 15.4 (6.7) burn
patients per year in contrast to 0.6 (0.6) burn patients
per year in general ICUs.
Table 1 shows the casemix of the 1141 admissions.
Median (IQR [range]) burn surface area was greater
for patients admitted to burn ICUs compared with
general ICUs (16 (7–32 [0–98]% vs. 8 (1–18 [0–
100]%), respectively; p ≤ 0.001). Admissions with asso-
ciated inhalation injury were similar for burn and
general ICUs (284/640 (44%) vs. 218/501 (44%),
respectively; p = 0.771). Not surprisingly, heat was the
most common burn mechanism. Overall, most admis-
sions came directly from the Emergency Department.
Patients were relatively young (mean age (SD) 47.8
(19.2) years) with almost three-quarters being men.
Most patients lived independently before admission
with few serious comorbidities reported.
In burn ICUs, organ support was received by a
higher proportion of patients, for longer duration and
both ICU and acute hospital durations of stay were
longer (Table 2). Almost one-third of burn patients
admitted to general ICUs were transferred to another
ICU. Crude mortality at ICU discharge was higher for
patients managed in burn ICUs (120/640 (19%) vs. 64/
501 (13%); p = 0.007). However, in-hospital mortality
was similar for both burn and general ICUs (146/640
(23%) vs. 95/501 (19%), respectively; p = 0.148). Most
of the non-survivors died in the ICU (Table 2).
For the comparison of the risk prediction models,
the analyses included the 1103 (97%) patients with
complete data for all variables for all models. Figure 2
presents the receiver operating characteristic curves for
the three models, Fig. 3 the calibration plots and
Table 3 presents and compares the measures of model
performance.

© 2018 The Association of Anaesthetists of Great Britain and Ireland 5

Anaesthesia 2018 Toft-Petersen et al. | Critical care organisation for burn patients in the UK

(a) Revised Baux model BOBI model ICNARC model

0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

Sensitivity

Sensitivity

Sensitivity
Ae
r a n
u e
d r O
R C c ru e
v = 0 8
. 6
7 4 Ae
r a n
u e
d r O
R C c ru e
v = 0 7
. 6
9 4 Ae
r a n
u e
d r O
R C c ru e
v = 0 9
. 7
2 6

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
1 - Specificity 1 - Specificity 1 - Specificity
(b) Revised Baux model BOBI model ICNARC model
0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

Sensitivity

Sensitivity

Sensitivity
Ae
r a n
u e
d r O
R C c ru e
v = 0 8
. 4
7 6 Ae
r a n
u e
d r O
R C c ru e
v = 0 7
. 0
8 9 Ae
r a n
u e
d r O
R C c ru e
v = 0 9
. 4
0 2

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
1 - Specificity 1 - Specificity 1 - Specificity

(c) Revised Baux model BOBI model ICNARC model

0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

0.00 0.25 0.50 0.75 1.00

Sensitivity

Sensitivity

Sensitivity

Ae
r a n
u e
d r O
R C c ru e
v = 0 8
. 8
7 3 Ae
r a n
u e
d r O
R C c ru e
v = 0 8
. 5
1 5 Ae
r a n
u e
d r O
R C c ru e
v = 0 9
. 7
5 1

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
1 - Specificity 1 - Specificity 1 - Specificity

Figure 2 Discrimination of three prediction models (revised Baux, Belgian Outcome in Burn Injury (BOBI) and
Intensive Care National Audit and Research Centre (ICNARC) model) for burned patients admitted to the intensive
care unit (ICU) in all complete cases. (a) all patients, (b) burn ICUs, (c) general ICUs. Patients were not included in
the risk prediction model comparisons if they had missing data for burn surface area, inhalation injury, acute physi-
ology score and/or outcome.

In burn ICUs, the revised Baux and BOBI risk BOBI models underestimated acute hospital mortality
prediction models had poorer discrimination than the in lower-risk patients and the ICNARC model in
ICNARC model, although this only reached signifi- higher-risk patients, yet the ICNARC model had
cance for the BOBI model. The revised Baux and slightly superior overall calibration. In general ICUs,

6 © 2018 The Association of Anaesthetists of Great Britain and Ireland

Toft-Petersen et al. | Critical care organisation for burn patients in the UK Anaesthesia 2018

(a) Revised Baux model BOBI model ICNARC model

100 100 100
Observed acute hospital

Observed acute hospital

Observed acute hospital

80 80 80
mortality (%)

mortality (%)

mortality (%)
60 60 60

40 40 40

20 20 20

20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

Model predicted Model predicted Model predicted
acute hospital mortality (%) acute hospital mortality (%) acute hospital mortality (%)

(b) Revised Baux model BOBI model ICNARC model

100 100 100
Observed acute hospital

Observed acute hospital

Observed acute hospital

80 80 80
mortality (%)

mortality (%)

mortality (%)
60 60 60

40 40 40

20 20 20

20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

Model predicted Model predicted Model predicted
acute hospital mortality (%) acute hospital mortality (%) acute hospital mortality (%)

(c) Revised Baux model BOBI model ICNARC model

100 100 100
Observed acute hospital

Observed acute hospital

Observed acute hospital

80 80 80
mortality (%)

mortality (%)

mortality (%)

60 60 60

40 40 40

20 20 20

20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

Model predicted Model predicted Model predicted
acute hospital mortality (%) acute hospital mortality (%) acute hospital mortality (%)

Figure 3 Calibration of three prediction models (revised Baux, Belgian Outcome in Burn Injury (BOBI) and Inten-
sive Care National Audit and Research Centre (ICNARC) model) for burned patients admitted to the intensive care
unit (ICU) in all complete cases. (a) all patients, (b) burn ICUs, (c) general ICUs. Patients were not included in the
risk prediction model comparisons if they had missing data for burn surface area, inhalation injury, acute physiology
score and/or outcome.

the overall discrimination of the ICNARC model was in lower-risk patients. The overall accuracy of the
excellent, whereas the burn-specific models were much ICNARC model was superior to that of the burn-
poorer. With respect to calibration, again, the burn- specific models, irrespective of the ICU setting.
specific models underestimated ultimate acute hospital With respect to the sensitivity analyses, exclusion
mortality, most markedly for lower-risk patients, of patients solely with electrical burns, or of patients
whereas the ICNARC model overestimated mortality in the original development sample for the ICNARC

© 2018 The Association of Anaesthetists of Great Britain and Ireland 7

Anaesthesia 2018 Toft-Petersen et al. | Critical care organisation for burn patients in the UK

Table 3 Indices of performance for the revised Baux (rBaux) and Belgian Outcome in Burn Injury (BOBI) and
Intensive Care National Audit and Research Centre (ICNARC) models. Patients were not included in the risk predic-
tion model comparisons if they had missing data for burn surface area, inhalation injury, acute physiology and/or
outcome.

Discrimination Calibration (Cox’s calibration regression)

Accuracy
p value for
p value vs. test of perfect
Model AUC (95%CI) ICNARC model Brier score Slope Intercept v2 calibration
All ICUs (n = 1103)
rBaux 0.88 (0.85–0.90) 0.00004 0.111 0.80 (0.69–0.91) 0.47 (0.21–0.73) 71.8 < 0.00001
BOBI 0.80 (0.77–0.83) < 0.00001 0.133 0.69 (0.58–0.80) 0.19 ( 0.05 to 0.43) 74.0 < 0.00001
ICNARC 0.93 (0.91–0.94) n/a 0.085 1.43 (1.24–1.61) 0.63 (0.34–0.92) 27.4 < 0.00001
Burn ICUs (n = 632)
rBaux 0.87 (0.84–0.90) 0.06546 0.118 0.81 (0.67–0.95) 0.39 (0.08–0.70) 30.9 < 0.00001
BOBI 0.78 (0.74–0.82) < 0.00001 0.146 0.64 (0.50–0.78) 0.14 ( 0.16 to 0.45) 51.3 < 0.00001
ICNARC 0.90 (0.88–0.93) n/a 0.098 1.40 (1.16–1.64) 0.77 (0.39–1.15) 18.8 0.00008
General ICUs (n = 471)
rBaux 0.88 (0.84–0.92) 0.00003 0.103 0.81 (0.64–0.98) 0.68 (0.21–1.14) 43.1 < 0.00001
BOBI 0.82 (0.77–0.86) < 0.00001 0.116 0.76 (0.57–0.95) 0.27 ( 0.14 to 0.68) 23.9 0.00001
ICNARC 0.96 (0.94–0.97) n/a 0.068 1.51 (1.19–1.83) 0.40 ( 0.05 to 0.84) 13.6 0.00114

AUC, area under the curve.

model, or recategorisation from general to burn ICU
for those transferred, did not alter the performance of
the risk prediction models (see Appendix S2).
Discussion
In accordance with UK policy, burn ICUs admitted
more severely injured patients (as indicated by higher
burn surface area) more frequently. Intensive care unit
mortality was higher in burn ICUs but in-hospital
mortality did not differ between burn and general
ICUs. The risk prediction models developed on burn
populations performed well, but had a poorer fit in a
burn patient ICU population compared with the
ICNARC risk prediction model that was developed in
a general critical care population.
The strengths of our study lie in the standardised,
accurate (complete, valid) data derived from a high-
quality clinical database covering a large population of
burn patients admitted to a large number of ICUs.
Almost 100% of general ICUs, but only 60% of burn
ICUs, were represented.
Comparisons across countries and healthcare set-
tings are difficult due to differences in the organisation
of care and how certain casemix or risk factors are
defined (e.g. a lack of a standardised definition for
inhalation injury [35]). In addition, few studies report
on burn patients in ICU, either in dedicated burn or
general ICUs. Although the age and sex distributions
are broadly similar across studies, the extent of injury
(burn surface area and the presence of inhalation
injury), as well as mortality (11–58%), differ vastly [18,
36–39].
Our finding of comparable mortality in burn and
general ICUs, in spite of patients admitted to burn
ICUs having suffered more extensive injuries, may
have different explanations. First, apart from the burn
surface area and the presence of inhalation injury, no
further clinical information regarding the injuries was
accessible. Thus, we do not know whether patients
who have suffered burns that are particularly prone to
become infected, (e.g. burns to the genital, perineal or
buttocks area [40, 41]), or burns that are accompanied
by other injuries, are more likely to be admitted to
either a burn or general ICU. Second, although the
ICNARC coding method captures severe comorbidity,
it does not capture milder pre-existing disease. It
might be that patients with mild pre-existing comor-
bidity are preferentially admitted to general ICU, and
as some burn patients stay in ICU for extensive peri-
ods of time, any such comorbidity may well influence
mortality. Third, it might be that burn ICUs provide
better care for patients suffering from this relatively
rare type of injury. However, the observational nature
of our data does not allow us to comment on the

8 © 2018 The Association of Anaesthetists of Great Britain and Ireland

Toft-Petersen et al. | Critical care organisation for burn patients in the UK
current quality of care provided for burn patients in
general ICUs and whether further efforts should be
made to admit patients primarily to burn ICUs. Given
that the UK policy is for burn patients to go to burn
ICUs, it is noteworthy that 30% of patients are trans-
ferred on to more specialised care, suggesting that
there may be limited capacity in burn ICUs or that
initial triage could be improved.
Since risk prediction models are important tools
for monitoring quality of care, it is paramount that
they are as fair and precise as possible. Ideally, a risk
prediction model should be generalisable across health-
care systems and different levels of care (ward to
ICU). The revised Baux and BOBI models, as well as
other models for risk prediction in burn patients, were
originally developed and later validated in populations
where the extent of injury was much lower, and they
are thus not necessarily accurate in ICU patient popu-
lations. As admittance to burn vs. general ICU might
depend on factors other than injury severity leading to
unpredictable selection, it is important to validate the
models in the types of units where they might be
employed. Few studies have validated the models in
ICUs.
A study of 48 patients admitted to a UK burn
ICU with a burn surface area higher than that in our
burn ICU population (30% vs. 23%), reported that the
APACHE-2 model was superior to the revised Baux
and BOBI models [42], and thus concurred with our
findings. In contrast, Tsurumi et al. [43] reported that
the revised Baux model had a better discrimination
than the APACHE-2 model in 333 burn patients in six
US burn ICUs. Only patients with a burn surface area
> 20% were included and the hospital mortality was
19%. In our study, the burn-specific models failed to
calibrate mainly in the lower-risk patients, which is to
be expected as patients admitted to ICU despite only
minor burns constitute a highly selected subset. Exclu-
sion of patients with burn surface area < 20% selects
higher risk populations (64% of our ICU patient popu-
lation would have been excluded had we applied this
criterion) and this may explain the discrepancy
between our findings and those of the above study.
The fit of burn-specific vs. generic risk predic-
tion models needs to be considered in the context
of the amount of data required to apply the model
© 2018 The Association of Anaesthetists of Great Britain and Ireland
Anaesthesia 2018
and the required accuracy for its use in clinical care.
Our finding that a generic risk prediction model
performs well in burn ICUs and outperforms burn-
specific models would suggest that benchmarking
and quality surveillance in burn ICUs could be
organised within the framework already set up for
general ICUs.
In conclusion, the UK policy of admitting burn
patients primarily to specialist ICUs is mostly com-
plied with. Although a number of burn patients are
cared for in general ICUs, the reasons for this need to
be investigated as issues of inadequate capacity may be
contributing. A generic risk prediction model performs
better in both burn and general ICUs compared with
commonly used burn-specific risk prediction models
for adults. Further work based on the whole adult and
paediatric burn population requiring ICU care, as well
as the non-ICU population, needs to be undertaken to
identify the optimal audit tools.
Acknowledgements
No external funding or competing interests declared.
References
1. World Health Organization. Burns. Fact sheet Numebr 365,
2018. http://www.who.int/mediacentre/factsheets/fs365/en/
(accessed 28/03/2018).
2. The National Burn Care Review Committee; British Burn Asso-
ciation. Standards and strategy for burn care: a review of burn
care in the British Isles. 2001. http://79.170.40.160/britishb
urnassociation.org/wp-content/uploads/2017/07/NBCR2001.
pdf (accessed 26/04/2018)
3. Brewster CT, Coyle B, Varma S. Trends in hospital admissions
for burns in England, 1991–2010: a descriptive population-
based study. Burns 2013; 39: 1526–34.
4. Regional Burns Service, Belfast Health and Social Care Trust.
http://www.belfasttrust.hscni.net/services/2623.htm (ac-
cessed 28/03/2018).
5. Stylianou N, Buchan I, Dunn KW. A review of the international
Burn Injury Database (iBID) for England and Wales: descriptive
analysis of burn injuries 2003–2011. BMJ Open 2015; 5:
e006184.
6. National Network for Burn Care. National Burn Care Referral
Guidance. 2012. http://www.britishburnassociation.org/down
loads/National_Burn_Care_Referral_Guidance_-_5.2.12.pdf
(accessed 10/01/2017).
7. Sheppard NN, Hemington-Gorse S, Shelley OP, Philp B,
Dziewulski P. Prognostic scoring systems in burns: a review.
Burns 2011; 37: 1288–95.
8. Jeschke MG, Pinto R, Kraft R, et al. Morbidity and survival
probability in burn patients in modern burn care. Critical Care
Medicine 2015; 43: 808–15.
9. El-Helbawy RH, Ghareeb FM. Inhalation injury as a prognostic
factor for mortality in burn patients. Annals of Burns and Fire
Disasters 2011; 24: 82–8.
9
Anaesthesia 2018 Toft-Petersen et al. | Critical care organisation for burn patients in the UK

rtega S, et al. A model for predicting
10. Galeiras R, Lorente JA, Pe
mortality among critically ill burn victims. Burns 2009; 35:
201–9.
11. Stylianou N, Buchan I, Dunn KW. A model of British in-
hospital mortality among burns patients. Burns 2014; 40:
1316–21.
12. Santaniello JM, Luchette FA, Esposito TJ, et al. Ten year experi-
ence of burn, trauma, and combined burn/trauma injuries
comparing outcomes. Journal of Trauma: Injury, Infection, and
Critical Care 2004; 57: 696–701.
13. Thombs BD, Singh VA, Halonen J, Diallo A, Milner SM. The
effects of preexisting medical comorbidities on mortality and
length of hospital stay in acute burn injury. Annals of Surgery
2007; 245: 629–34.
14. Heng JS, Clancy O, Atkins J, et al. Revised Baux Score and
updated Charlson comorbidity index are independently associ-
ated with mortality in burns intensive care patients. Burns
2015; 41: 1420–7.
15. Dokter J, Meijs J, Oen IMMH, van Baar ME, van der Vlies CH,
Boxma H. External validation of the revised Baux score for the
prediction of mortality in patients with acute burn injury.
Journal of Trauma and Acute Care Surgery 2014; 76: 840–5.
16. Haum A, Perbix W, H€ack HJ, Stark GB, Spilker G, Doehn M.
Alcohol and drug abuse in burn injuries. Burns 1995; 21:
194–9.
17. Moore EC, Pilcher DV, Bailey MJ, Cleland H, McNamee J. A sim-
ple tool for mortality prediction in burns patients: APACHE III
score and FTSA. Burns 2010; 36: 1086–91.
18. Queiroz LFT, Anami EHT, Zampar EF, Tanita MT, Cardoso LTQ,
Grion CMC. Epidemiology and outcome analysis of burn
patients admitted to an Intensive Care Unit in a University
Hospital. Burns 2016; 42: 655–62.
19. Pavoni V, Gianesello L, Paparella L, Buoninsegni L, Barboni E.
Outcome predictors and quality of life of severe burn patients
admitted to intensive care unit. Scandinavian Journal of
Trauma, Resuscitation and Emergency Medicine 2010; 18: 24.
20. O’Keefe GE, Hunt JL, Purdue GF. An evaluation of risk factors
for mortality after burn trauma and the identification of gen-
der-dependent differences in outcomes. Journal of the Ameri-
can College of Surgeons 2001; 192: 153–60.
21. McGwin G, George RL, Cross JM, Reiff DA, Chaudry IH, Rue LW.
Gender differences in mortality following burn injury. Shock
2002; 18: 311–15.
22. Moore EC, Pilcher D, Bailey M, Cleland H. Women are more
than twice as likely to die from burns as men in Australia
and New Zealand: an unexpected finding of the Burns Evalua-
tion And Mortality (BEAM) Study. Journal of Critical Care 2014;
29: 594–8.
23. Ray JJ, Satahoo SS, Meizoso JP, et al. Does obesity affect out-
comes of adult burn patients? Journal of Surgical Research
2015; 198: 450–5.
24. Hussain A, Choukairi F, Dunn K. Predicting survival in thermal
injury: a systematic review of methodology of composite pre-
diction models. Burns 2013; 39: 835–50.
25. Harrison DA, Brady AR, Rowan K. Case mix, outcome and
length of stay for admissions to adult, general critical care
units in England, Wales and Northern Ireland: the Intensive
Care National Audit & Research Centre Case Mix Programme
Database. Critical Care 2004; 8: R99–111.
26. Young JD, Goldfrad C, Rowan K. Development and testing of a
hierarchical method to code the reason for admission to
intensive care units: the ICNARC coding method. British Jour-
nal of Anaesthesia 2001; 87: 543–8.
10
27. Harrison DA, Parry GJ, Carpenter JR, Short A, Rowan K. A new
risk prediction model for critical care: the Intensive Care
National Audit & Research Centre (ICNARC) model. Critical
Care Medicine 2007; 35: 1091–8.
28. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a
severity of disease classification system. Critical Care Medicine
1985; 13: 818–29.
29. Information Standards Board for Health and Social Care. Criti-
cal care minimum dataset full specification critical care mini-
mum dataset version 8.0. 2010. http://webarchive.nationala
rchives.gov.uk/+/http://www.isb.nhs.uk/documents/isb-0153/
amd-81-2010/0153812010spec.pdf (accessed 28/03/2018).
30. Osler T, Glance LG, Hosmer DW. Simplified estimates of the
probability of death after burn injuries: extending and updat-
ing the Baux score. Journal of Trauma: Injury, Infection, and
Critical Care 2010; 68: 690–7.
31. The Belgian Outcome in Burn Injury Study Group. Develop-
ment and validation of a model for prediction of mortality in
patients with acute burn injury. British Journal of Surgery
2009; 96: 111–17.
32. Ferrando-Vivas P, Jones A, Rowan KM, Harrison DA. Develop-
ment and validation of the new ICNARC model for prediction
of acute hospital mortality in adult critical care. Journal of
Critical Care 2016; 38: 335–9.
33. Cox DR. Two further applications of a model for binary regres-
sion. Biometrika 1958; 45: 562–5.
34. Koumbourlis AC. Electrical injuries. Critical Care Medicine
2002; 30: S424–30.
35. Walker PF, Buehner MF, Wood LA, et al. Diagnosis and man-
agement of inhalation injury: an updated review. Critical Care
2015; 19: 351.
36. McNamee JJ, Pilcher DV, Bailey MJ, Moore EC, Cleland HJ. Mor-
tality prediction and outcomes among burns patients from
Australian and New Zealand intensive care units. Critical Care
and Resuscitation 2010; 12: 196–201.
37. Moore EC, Pilcher DV, Bailey MJ, Stephens H, Cleland H. The
Burns Evaluation and Mortality Study (BEAMS): predicting
deaths in Australian and New Zealand burn patients admitted
to intensive care with burns. Journal of Trauma and Acute
Care Surgery 2013; 75: 298–303.
38. de Campos EV, Park M, Gomez DS, Ferreira MC, Azevedo
LCP. Characterization of critically ill adult burn patients admit-
ted to a Brazilian intensive care unit. Burns 2014; 40:
1770–9.
39. Ibarra Estrada MA,
Ch
avez Pen ~a Q, Garc
ıa Guardado DI, Lo
pez
Pulgar
ın JA, Aguirre Avalos G, Corona JF. A 10-year experience
with major burns from a non-burn intensive care unit. Burns
2014; 40: 1225–31.
40. Harpole BG, Wibbenmeyer LA, Erickson BA. Genital burns
in the national burn repository: incidence, etiology, and
impact on morbidity and mortality. Urology 2014; 83:
298–302.
41. Clemens MS, Janak JC, Rizzo JA, et al. Burns to the genitalia,
perineum, and buttocks increase the risk of death among U.S.
service members sustaining combat-related burns in Iraq and
Afghanistan. Burns 2017; 43: 1120–8.
42. Douglas HE, Ratcliffe A, Sandhu R, Anwar U. Comparison of
mortality prediction models in burns ICU patients in Pinder-
fields Hospital over 3 years. Burns 2015; 41: 49–52.
43. Tsurumi A, Que Y-A, Yan S, Tompkins RG, Rahme LG, Ryan
CM. Do standard burn mortality formulae work on a popula-
tion of severely burned children and adults? Burns 2015; 41:
935–45.
© 2018 The Association of Anaesthetists of Great Britain and Ireland
Toft-Petersen et al. | Critical care organisation for burn patients in the UK Anaesthesia 2018

Supporting Information Appendix S1. The risk prediction models.

Additional Supporting Information may be found in Appendix S2. Sensitivity analyses.
the online version of this article:

© 2018 The Association of Anaesthetists of Great Britain and Ireland 11