Developing A Burn Injury Severity Score

JBUR-4186; No.
of Pages 9
burns xxx (2013) xxx–xxx
Available online at www.sciencedirect.com
ScienceDirect
journal homepage: www.elsevier.com/locate/burns
Developing a burn injury severity score (BISS):

Adding age and total body surface area burned to
the injury severity score (ISS) improves mortality
concordance
J. Tristan Cassidy a, Michael Phillips b, Daniel Fatovich c, Janine Duke d,

Dale Edgar e, Fiona Wood f,*
a
Royal Perth Hospital, Australia
b
Western Australian Institute for Medical Research, University of Western Australia, Australia
c
Emergency Medicine, Royal Perth Hospital, University of Western Australia, Australia
d
Epidemiology, Burn Injury Research Unit, University of Western Australia, Australia
e
Fiona Wood Foundation, Royal Perth Hospital, Australia
f
University of Western Australia, Australia
article info abstract
Article history: Background: There is limited research validating the injury severity score (ISS) in burns. We
Accepted 13 October 2013 examined the concordance of ISS with burn mortality. We hypothesized that combining age
and total body surface area (TBSA) burned to the ISS gives a more accurate mortality risk
Keywords: estimate.
Burn Methods: Data from the Royal Perth Hospital Trauma Registry and the Royal Perth Hospital
Trauma Burns Minimum Data Set were linked. Area under the receiver operating characteristic
Mortality curve (AUC) measured concordance of ISS with mortality. Using logistic regression models
Injury severity score with death as the dependent variable we developed a burn-specific injury severity score
ISS (BISS).
Results: There were 1344 burns with 24 (1.8%) deaths, median TBSA 5% (IQR 2–10), and
median age 36 years (IQR 23–50). The results show ISS is a good predictor of death for burns
when ISS 15 (OR 1.29, p = 0.02), but not for ISS > 15 (ISS 16–24: OR 1.09, p = 0.81; ISS 25–49:
OR 0.81, p = 0.19). Comparing the AUCs adjusted for age, gender and cause, ISS of 84% (95% CI
82–85%) and BISS of 95% (95% CI 92–98%), demonstrated superior performance of BISS as a
mortality predictor for burns.
Conclusion: ISS is a poor predictor of death in severe burns. The BISS combines ISS with age
and TBSA and performs significantly better than the ISS.
# 2013 Elsevier Ltd and ISBI. All rights reserved.
* Corresponding author at: Royal Perth Hospital, 197 Wellington Street, Perth, Western Australia, Australia. Tel.: +61 892243566;
fax: +61 892243577.
E-mail addresses: Tristancassidy@gmail.com (J.T. Cassidy), Michael.phillips@waimr.uwa.edu.au (M. Phillips),
Daniel.Fatovich@health.wa.gov.au (D. Fatovich), Janine.Duke@uwa.edu.au (J. Duke), Dale.Edgar@health.wa.gov.au (D. Edgar),
Fiona.Wood@health.wa.gov.au (F. Wood).
0305-4179/$36.00 # 2013 Elsevier Ltd and ISBI. All rights reserved.
http://dx.doi.org/10.1016/j.burns.2013.10.010
Please cite this article in press as: Cassidy JT, et al. Developing a burn injury severity score (BISS): Adding age and total body surface area burned
to the injury severity score (ISS) improves mortality concordance. Burns (2013), http://dx.doi.org/10.1016/j.burns.2013.10.010
JBUR-4186; No. of Pages 9
2 burns xxx (2013) xxx–xxx
upon a coding lexicon originally developed by the Association

1. Background for the Advancement of Automotive medicine, known as the
Abbreviated Injury Scale (AIS) [4,10]. The AIS grades burn
Burns are a significant cause of trauma mortality and a major severity based on the percentage of total body surface area
public health problem globally [1]. A systematic burn (TBSA) burned and does not discriminate between 2nd degree
epidemiology review of European data reported mortality and 3rd degree burns above 10% TBSA. This is outlined in Table
ranging from 1.4 to 18% for severe burn (where ‘‘severe burn’’ 1. Inhalation injury is coded separately using five grades with
was defined as an acute burn in need of specialized care during corresponding ISS scores ranging from 4 to 75 and is outlined
hospital admission) [2]. A ten year review of the U.S. national in Table 2. Based on this grading system a 3rd degree burn of
burn registry reported mortality of 5.6% [3]. While patients 80% TBSA will be coded the same as a 40% TBSA 2nd degree
with severe burns are predominately cared for in specialized burn. For an isolated burn, that is a burn with no other
burn centers, for the majority of severe burns not treated at concomitant traumatic injuries, these two burn scenarios will
such burn centers, management is provided at other high have the same ISS.
volume hospitals (i.e. greater than 100 burn admissions per The relationships between TBSA, age and burn mortality
year), e.g. trauma centers [4,5]. Consequentially, burns are a are well described [6,11]. Our clinical observation is that the ISS
common presentation to trauma centers [6,7]. is a less reliable predictor of death in severe burns. This paper
Scoring systems are used to predict risk of death after aims to examine the concordance of ISS with burn mortality.
trauma. The most popular trauma scoring system is the injury We hypothesize that combining age and TBSA as continuous
severity score (ISS) [8,9]. However there is little research variables to the ISS will give a more accurate mortality risk
validating the ISS in burns. The ISS categorizes burn based estimate.
Table 1 – The Abbreviated Injury Scale (AIS) coding system for burns.b
Coding Description Specifications TBSA % AIS ISS
1st degree; superficial if >1 yo Any 1 1
1 yo 50% 1 1
<1 yo >50% 2 4
2nd degree; partial thickness <10% 1 1

2 2
3rd degree; full thickness 100 cm [face < 25 cm ] 1 1
>100 cm2 < 10% [face > 25 cm2] 2 4
2nd or 3rd degree; partial or full thickness 10–19% 2 4

<5 yo 3 9

<5 yo 4 16

<5 yo 5 25

2nd or 3rd degree; partial or full thickness including incineration 90% 6 75a
AIS = Abbreviated Injury Scale; ISS = injury severity score; TBSA = total body surface area burned.
a
Any injury coded as a 6 will automatically give an ISS of 75.
b
When burns occur in varying degrees the rater assigns an AIS code to the first degree burns separately from second and third degree Burns. If
second degree burns are <10% TBSA and/or third degree burns <100 cm2 or >100 cm2 but <10%, then both second degree and third are
combined to give a single score.
Table 2 – The Abbreviated Injury Scale (AIS) coding system for inhalation injury.b
Coding description AIS ISS
Inhalation injury NFS (heat, particulate matter, caustic or noxious agents) 2 4
Absence of carbonaceous deposits, erythema, edema, bronchorrhea, obstruction 2 4
Minor or patchy areas of erythema, bronchorrhea, carbonaceous deposits in proximal or distal bronchi 3 9
Moderate degree of erythema, carbonaceous deposits, bronchorrhea with or without compromise of the bronchi 4 16
Severe inflammation with friability, copious carbonaceous deposits, bronchorrhea, bronchial obstruction, hypoxemia 5 25
Evidence of mucosal sloughing, necrosis, endoluminal obliteration 6 75a
AIS = Abbreviated Injury Scale; ISS = injury severity score; NFS = not further specified.
a
Any injury coded as a 6 will automatically give an ISS of 75.
b
Includes all airway burns from mouth and nose to lungs – mouth or pharynx not coded separately.
burns xxx (2013) xxx–xxx 3
(1) An examination of the concordance of the ISS with

2. Methods observed probability of death by mechanism of injury.
(a) The AUC was generated and ranked for each cause
2.1. Study design group.
(b) Confounders/potential indicators of severity were
This is a single center, adult hospital-based retrospective identified using logistic regression. Other potential
study using clinical data that have been prospectively indicators of severity were considered but rejected
collected. Study data were obtained from the Royal Perth either because of an excessive number of missing
Hospital Trauma Registry (RPHTR) and the Royal Perth values (including burn agent) or because of timing
Hospital Burns Minimum Data Set (RPHBMDS) with approval limitations affecting clinical utility (including: number
from Royal Perth Hospital Human Research Ethics Committee of surgical procedures, length of stay, admission to
(Ethics Approval number EC 2012/127). The RPHTR began intensive care unit). Due to the absence of consistent
prospective data collection in August 1994 for the Western diagnostic criteria, inhalation injury was not consid-
Australia State Major Trauma Service at Royal Perth Hospital ered.
(RPH), a verified Level 1 Trauma center. The RPH Burns (c) Following identification of age and gender as con-
Minimum Data Set was established by the Burns Service WA in founders the AUC values were recalculated after
2004 as an internationally recognized minimum data set that adjustment. This analysis was limited to causes with
includes patient and clinical burn management data. The at least 5 deaths. The ‘‘other’’ cause group was
Burns Unit at RPH maintains the RPHBMDS on a prospective excluded from this analysis due to its heterogeneous
basis. All adult patients hospitalized for a burn are included in nature.
this data set. Pediatric cases are not included and were thus (d) The ROC curve was generated comparing burn with
excluded from this study. other causes. Detailed analysis of the AUC within
These two data sets were linked using a unique medical specific regions of the ROC curve using the partial AUC
record number and date of injury and de-identified prior to was considered but did not give interpretable results.
supply to researchers. Data for the period 2004–2011 inclusive (e) Using the severity categories described by Sampalis
were used in this study. Mortality was the primary outcome odds ratios for each cause group were calculated and
measure and was defined as a record of death that occurred compared.
after presentation to the Emergency Department during the (2) The development of a burn-specific injury severity score
index hospital admission for the respective trauma-related (BISS).
injury (burn, non-burn). Only index trauma cases were (a) This used logistic regression models with death as the
analyzed. The variables accessed in these data sets included dependent variable and the severity indicators together
age, gender, mechanism of injury, TBSA, inhalation injury, with potential effect modifiers of mortality risk
burn depth, concomitant injury, ISS and mortality. Cause particularly age, gender and TBSA.
of trauma was classified according to mechanism of injury. (b) Gender was not a significant independent mortality
ISS was categorized using the Sampalis classification: ISS 16– predictor for the burns cohort, nor did it interact with
24 (survivable), ISS 25–49 (probably survivable) and ISS 50+ any significant predictors, and was thus excluded from
(Non-Survivable) and with the addition of ISS 0–15 (minor) BISS.
[8,12]. (c) Interaction terms and non-linear transformations
using fractional polynomials were incorporated in
2.2. Statistical analysis the model development. Interaction significance was
established using a likelihood ratio test.
All analyses were conducted using the Stata V12.1 statistical (d) The AUC was used to identify models with high
software (StataCorp LP, College Station, TX). A p value less discrimination. The AIC was used to compare models.
than 0.05 was considered to be statistically significant. (e) ROC curves and AUC values with confidence intervals
Proportions, medians and inter-quartile range were used to were generated to compare concordance of both the ISS
describe the sample. Inferential methods primarily used and BISS with mortality.
logistic regression models with death as the dependent (f) Curves of agreement between ISS and BISS vs. observed
variable. Graphs for observed probability of death by ISS were risk of death were generated. These demonstrated
generated using the Lowess smoothing method. Area under where the estimated risk of death for each model
the receiver operating characteristic (ROC) curve (AUC) was agreed and differed.
used to identify models with high discrimination. ROC (3) Validation of the new index and limitation of over-fitting
regression methods of Pepe were used to compare indicators (a) This was based upon a bootstrapped logistic regression
of mortality risk [10,13,14]. Models were compared using the model with death as the outcome and BISS as the only
Akaike information criterion (AIC) [11]. The linearity assump- independent variable. Bootstrapping is a statistical
tion for the logistic regression models was tested using method of re-sampling which avoids making assump-
goodness of fit (GOFHL) tests by the method of Hosmer and tions about the distributions within the dataset. Similar
Lemeshow [12]. The GOFHL tests were adjusted for multiple models were run with ISS as the independent variable
comparisons using the Holm–Simes method of sequential for comparison purposes. The bootstrapped models
rejection [13,14]. were run with 1000 repetitions and the bias-corrected
The strategy for the analysis followed these steps: confidence intervals are reported in Section 3.
Table 3 – Outline of burn deaths.

Cause Gender Cause of death Age ISS TBSA%
Fire Male Resp fail 66 1 2
Fire Female Resp fail 65 10 2
Fire Male MOF 73 4 13
Fire Male CVA 80 4 15
Fire Male Unknown 43 4 15
Fire Female Pneumonia 67 18 20
Fire Female Sepsis 67 16 25
Fire Female Pneumonia 86 9 25
Fire Male Cardio event 93 25 31
Explosion Female IHD 70 26 35
Fire Female MOF 56 25 40
Fire Female Sepsis 58 25 45
Fire Male Sepsis 81 25 50
Fire Male Pneumonia 28 25 75
Firea Male Un-survivable 48 75 87.5
Firea Male Un-survivable 45 75 96
Explosion Male Brain injury 28 26 Unknown
ISS = Injury severity Score; TBSA = % of total body surface area burned; Resp fail = respiratory failure.
a
Un-survivable inhalation injury as categorized by AIS (Abbreviated Injury Scale) coding.
median age was 36 years (IQR 23–50) and median TBSA was 5%
3. Results (IQR 2–10); TBSA values were missing for 102 (7.6%) burn
patients, and one of the burn deaths. Injury descriptions
The total number of patients included in this study from the showed that some of the missing TBSA values concerned
RPHTR data was 34,572 with linked data from the BMDS for a patients with inhalation injury but no external burn. Char-
nested cohort of 1344 (3.9%) index burn patients. This was the acteristics of the 24 (1.8%) burn deaths are presented in Table 3.
total number of patients captured by the BMDS from com- Inhalation injury was coded as suspected in seven of the burn
mencement of data collection in 2004 to the end of 2011. Of the deaths and was confirmed bronchoscopically in two patients.
total study sample 22,836 (66.1%) were male, the median age Concomitant injury was recorded in four patients.
was 41 years (IQR 26–65) and the median ISS was 4 (IQR 1–9). There were 33 trauma patients with a TBSA 40% for
There were 672 (1.9%) deaths. For the 1344 burn cases, the whom the TBSA ranged from 40 to 90% with a median of 50%
Table 4 – Concordance between injury severity score and the observed probability of death as measured by Area under
ROC Curve for causes where deaths >5.a
Cause of trauma Number of Number of Observed probability AUC % 95% CI for Rank of
cases (%) deaths of deatha AUC AUCb
LCL UCL
Motor cycle crash 2745 (7.9) 42 0.0153 97 96 98 1
Other 1611 (4.7) 29 0.0178 97 93 99 2
Cyclist/Pedestrian 1466 (4.2) 46 0.0314 95 93 97 3
Motor vehicle crash 3824 (11.1) 111 0.0290 90 87 93 4
Burn 1344 (3.9) 24 0.0179 90 82 95 5
Struck by object 3850 (11.1) 20 0.0052 89 78 96 6
Gunshot/stabbing 525 (1.5) 23 0.0438 89 79 94 7
Fall from height 2999 (8.7) 59 0.0197 85 79 91 8
Fall from standing 9091 (26.3) 307 0.0338 76 73 79 9
All causes (deaths >5) 34,572 673 0.0194 85 83 87
Note: For causes of trauma with more than 5 deaths (7123 cases and 11 deaths excluded). AUC = area under receiver operating characteristic
(ROC) curve; ISS = injury severity score.
a
Death occurring after arrival at ED and before discharge.
b
Rank by AUC and probability of death for ties.
Table 5 – Age and gender adjusted area under ROC curve for injury severity score by cause of injury where number of
deaths >5.a
Trauma cause Gender coefficient ( p) Age coefficient ( p) Cause coefficient ( p) Adjusted AUC % 95% CI AUC
Burn 0.35 (<0.001) 0.0075 (<0.001) 2.81 (<0.001) 84 82–85
Struck by object 0.36 (<0.001) 0.0055 (0.002) 1.32 (<0.001) 84 82–85
Fall from Standing 0.20 (0.026) 0.0166 (<0.001) 0.74 (<0.001) 84 82–86
Gunshot/Stabbing 0.31 (<0.001) 0.0088 (<0.001) 0.36 (0.27) 83 81–85
Fall from Height 0.29 (0.001) 0.0075 (<0.001) 1.80 (<0.001) 84 82–85
Cyclist/Pedestrian 0.30 (0.001) 0.0100 (<0.001) 3.23 (<0.001) 83 82–85
Motor cycle crash 0.04 (0.69) 0.0151 (<0.001) 4.39 (<0.001) 84 82–85
Motor vehicle crash 0.68 (<0.001) 0.0221 (<0.001) 6.10 (<0.001) 83 81–85
AUC = area under receiver operating characteristic curve.
a
(IQR 45–70%). These consisted of both isolated burns and Odds ratios for severity categories defined by Sampalis are
burns with concomitant trauma [6]. Of these, the primary shown in Table 6. For minor severity (ISS < 15), the Burn and
causes of trauma recorded were 23 (69.7%) Fire, 3 (9.1%) Fall from standing categories have a statistically significant
Explosion and 2 (6.1%) Chemical, with single cases (3.0%) for odds ratio for the ISS. This indicates that ISS is a good predictor
Electrical, Motor Cycle Crash (MCC), Motor Vehicle Crash of the risk of death for burns classified as minor trauma with 7
(MVC), Radiant Heat and Struck by Object (SBO). The (0.6%) burn deaths (OR 1.29, 95% CI 1.05–1.6, p = 0.02). For the
corresponding median ISS was 25 (IQR 25–26). Twenty-one moderate group (ISS 16–24) only the MVC odds ratio was
(63.6%) of these 33 patients had an ISS of 25. Within the significant. ISS was statistically significant for the MVC, MCC,
subset of patients with ISS of 25, the TBSA ranged from 40 to Cyclist/Pedestrian and SBO groups for the severe category (ISS
85%. There were 11 burn deaths with TBSA 40%, three of 25–49). For the ISS 50+ group only the MVC group was
which had TBSA of 90%. The ISS values in the remaining 8 significant. This pattern confirms the contrast in Fig. 1.
deaths with TBSA 40% were 25 in six cases, 29 in one and 75 The burn injury severity score (BISS) was developed by
in the final case. examination of ISS and other variables thought to be
Table 4 ranks the causes of trauma according to the ability associated with severity of injury based upon clinical
of the ISS to predict death as indicated by the AUC. It facilitates judgment and previous research. Having rejected burn depth
comparison of the concordance of the AUC for death between and gender, BISS was computed by modeling death with TBSA
causes. Across all causes of trauma the concordance of ISS was
85% (95% CI 83–87%). For specific causes with deaths greater
than 5, the concordance ranges from 76% to 97%. Burn ranked
5th when considering only causes with greater than 5 deaths.
It had an AUC of 90% (95% CI 82–95%).
Table 5 shows the AUC for ISS adjusted for the influence of
age, gender and cause. Adjustment for age and gender was
necessary because they were identified as significant con-
founders and were heterogeneous between cause groups. No
other significant confounders were identified. The AUC for
the different causes is very similar following adjustment. The
estimates ranged from 83% to 84% with a weighted mean of
83% for all causes. The coefficients represent the margin of
adjustment for each cause, by which the probability of death
should be increased from the baseline trauma mortality risk.
The coefficient for the cause is significant for all causes
except Gunshot/Stabbing. After age and gender adjustment
the coefficients range from 2.81 for burns to 6.10 for MVCs.
The negative cause coefficient for burn is contrasted with the
positive coefficient for motor vehicle crash injuries using
ROC curves in Fig. 1. These two causes have almost identical
crude and age–gender adjusted AUCs. The figure demon-
strates that ISS for burns has higher sensitivity when
specificity is high and death is unlikely, but lower sensitivity Fig. 1 – ROC curve analysis representing the concordance
as specificity decreases and death becomes more likely in with death of the injury severity score for burns and motor
contrast to the MVC ROC curve. This results in the two ROC vehicle crashes. With decreasing specificity (or increasing
curves crossing. Given this unusual ROC curve for burns it 1-specificity as plotted above) the ISS is relatively less
was decided to examine the performance of the ISS by degree sensitive for burns when compared to motor vehicle
of severity. crashes.
Table 6 – Risk of death associated with injury severity score (odds ratio) for each Sampalis injury severity score category
for causes where deaths >5.a
Cause ISS category Odds ratio 95% CI for OR p Number of cases Number of deaths (%)
Burn 0–15 1.29 1.05–1.59 0.02 1252 7 (0.6)
16–24 1.09 0.54–2.18 0.81 40 2 (5.0)
25–49 0.81 0.58–1.12 0.19 47 11 (23.4)
50+ ne 1 ne ne 5 5 (100)
Fall from height 0–15 1.12 0.98–1.28 0.10 2617 17 (0.7)
16–24 1.07 0.59–1.94 0.83 194 2 (1.0)
25–49 1.04 0.97–1.11 0.97 178 36 (20.2)
50+ 1.07 0.91–1.26 0.39 10 4 (40.0)
Fall from standing 0–15 1.22 1.16–1.28 <0.01 8660 214 (2.5)
16–24 0.98 0.71–1.31 0.83 208 23 (11.1)
25–49 1.12 0.96–1.26 0.15 223 70 (31.4)
50+ ne 2 ne ne 0
Gunshot/stabbing 0–15 1.16 0.88–1.53 0.29 436 3 (0.7)
16–24 0.73 0.48–1.12 0.15 51 10 (19.6)
25–49 0.84 0.55–1.30 0.41 38 10 (26.3)
50+ ne 2 ne ne 0
Motor vehicle crash 0–15 1.10 0.93–1.29 0.26 2708 8 (0.3)
16–24 1.25 1.04–1.51 0.02 565 20 (3.5)
25–50 1.08 1.04–1.13 <0.01 495 58 (11.7)
50+ 1.08 1.01–1.15 0.02 56 25 (44.6)
Motor cycle crash 0–15 ne 3 ne ne 2167 0 (0.0)
16–24 0.96 0.62–1.48 0.86 305 3 (1.0)
25–49 1.13 1.06–1.21 <0.01 245 27 (11.0)
50+ 1.08 0.99–1.17 0.07 28 12 (42.9)
Cyclist/pedestrian 0–15 1.46 0.77–2.77 0.24 1153 1 (0.1)
16–24 1.01 0.76–1.33 0.96 164 8 (4.9)
25–49 1.11 1.04–1.19 <0.01 139 33 (23.7)
50+ 1.07 0.93–1.23 0.34 10 4 (40.0)
Struck by object 0–15 1.15 0.94–1.43 0.18 3536 5 (0.1)
16–24 0.99 0.59–1.66 0.96 210 4 (1.9)
25–49 1.12 1.01–1.24 0.03 102 10 (9.8)
50+ ne 2 ne ne 2 1 (50.0)
Total 25,844a 633 (2.4)

1 2
ISS = injury severity score, CI = confidence interval. ne = not estimable because none survived. ne = not estimable because of insufficient
numbers. ne3 = not estimable because all survived.
a
and ISS age as an interaction term. The logistic regression the region of improved performance of BISS occurs when the
equation is: probability of death is higher, i.e. more severe burn.
BISS ¼ expð6:34 þ 0:0031 ðISS ageÞ þ 0:033 TBSAÞ: The GOFHL and bootstrapping of the regression model
analyzed for over-fitting. The GOFHL p-value for BISS was 0.77
The BISS model was evaluated using the AIC and the and for ISS it was 0.27. Both of which were acceptable. A test of
Hosmer and Lemeshow GOFHL test based upon bootstrapped the equality of the AUC between the two predictors showed
models to control for over-fitting [1,7,11]. Given any two that they were not equal ( p = 0.005) and post hoc tests of
estimated models, the model with the lower value of AIC is equality showed that BISS had a significantly greater AUC than
preferred. The AIC value for ISS was 5313 and for BISS it ISS ( p = 0.016).
was 129. As an added check of the robustness of the BISS model,
The ROC curves for BISS and ISS are shown in Fig. 2. The logistic regression modeling was performed on the sample
BISS demonstrates an AUC of 95% (CI 92–98%) vs. the ISS (age excluding the two burn deaths where TBSA was 2% (Table 3).
and gender adjusted) AUC of 84% (95% CI 82–85%). Including a This analysis demonstrated that the BISS is sensitive to these
four-category indicator of burn depth improved the BISS AUC two cases. BISS has a greater AUC (0.972 vs. 0.954), a lower
to 97% but all terms became non-significant in a boot- AIC (87.3 vs. 129) and maintained a non-significant GOFHL
strapped model indicating that the model was over-fitted and when the two burns deaths with 2% TBSA were excluded.
thus not valid. The median BISS value was 63 (IQR 36–197) for These results indicate that the BISS works better when only
all burns. burns with an independent mortality risk are included.
Fig. 3 illustrates the comparative performance of the BISS However, with the exclusion of these two burns deaths of 2%
and the ISS as predictors of the probability of death. Where TBSA, bootstrap analysis demonstrated this model to be
probability of death is low, the BISS and ISS agree closely but over-fitted which might have been predicted with a concor-
with increasing probability of death BISS and ISS diverge. Thus dance of 97%.
with burn mortality risk compared with all causes (AUC 85%).
However, following age and gender adjustment the respective
AUCs for each cause category were found to be remarkably
similar. The AUCs ranged from 83 to 84%, with a weighted
mean of 83%. Review of burns with TBSAs 40% in our dataset
highlighted a wide range of TBSAs and associated increased
mortality risk, but minimal variability in the ISS score. The
negative cause co-efficient for burns (Table 5) implies a
reduced risk of death with increasing severity. Therefore the
ISS performs poorly as a valid predictor of death as severity
increases.
Close examination of the ISS ROC curve for burns revealed
an atypical shape with sensitivity being high when specificity
is high and vice versa. The odds ratios for burns, consistent
with the negative cause coefficient, fall with increasing
severity even going below 1.00 in the ISS 25–49 group and
not calculable in the ISS 50+ group (where all patients died).
For burns, the only significant odds ratio was for the least
severe injury group. The MVC group (for which the ISS is a far
more established mortality prediction tool) contrasts with this
by having no odds ratio below 1.00 and having statistical
significance in the higher severity categories.
Fig. 2 – ROC curve analysis representing the concordance Note the far greater number of patients within the minor
with death of the burn injury severity score (BISS) and the category for burns. The high proportion of burn subjects
injury severity score (ISS). For burns, the BISS has an within this category, the only category where the ISS performs
improved ROC curve and associated increased Area Under well as a predictor of death, accounts for the misleadingly
the ROC curve (AUC) vs. the ISS (95% vs. 85%). high AUC value for burns. All of our results indicate that the
higher the risk of death as a result of burn, the poorer the
performance of the ISS.
The over-arching objective of the study was, therefore, to
develop a solution to the problems identified with the ISS as a
mortality predictor after burn. Thus, the subsequent analysis
leading to the development of the BISS. The BISS model
includes age, TBSA and ISS as continuous variables instead of
the TBSA categories used to calculate ISS (Table 1). It might be
considered that by introducing TBSA as a separate variable, in
addition to its incorporation within the ISS, there may be
concern that the BISS uses TBSA twice in contradiction of the
usual practice in statistical methodology of hypothesis testing.
However, the BISS is not designed to test hypotheses of causal
relationships for burn death, but rather as a valid statistical
indicator of mortality risk for use in observational research. In
BISS we have reported the combination of age, TBSA and ISS
which gives the most accurate mortality estimate (based upon
our data).
A limitation of this study was that all data were derived
from a single center. However, the data were collected from
two independent sources. The sample size was large, but these
Fig. 3 – Agreement of burn injury severity score (BISS) and are consecutive cases and the event rate was low. Two cases
injury severity score (ISS) in prediction of the probability of included as burn deaths in our data had 2% TBSA. Although it
burn death. There is increasing disagreement between the is likely that these deaths involved factors associated with co-
BISS vs. the ISS with increasing probability of death. morbidity (no other concomitant injuries were recorded for
these cases) they were included in the analysis that generated
the BISS because there were no prior grounds for their
exclusion.
4. Discussion The sensitivity analysis demonstrated a bias that shows
that BISS underestimates the risk of death. Despite this
The study confirmed that the ISS is a poor predictor of post- limitation it is the authors’ opinion that the BISS as derived
burn mortality as burn severity increases. Our initial analysis from the ISS has the ability to incorporate factors related to
in Table 4 found that with an AUC of 90%, the ISS concords well burn death risk such as concomitant injury and burn depth.
Hence, the BISS has the potential to apply to burn cases where conception and design, statistical analysis, data interpretation
both the burn and other trauma are factors affect mortality and paper composition. Daniel Fatovich contributed to study
risk. design, data interpretation, paper composition and critical
Another limitation of our study is that 102 subjects (7.6%) revision. Janine Duke contributed to study design, data
had missing TBSA values. We believe that most of these are interpretation, paper composition and critical revision. Dale
minor burns or those that are limited to inhalation injury Edgar contributed to study conception, design and critical
based upon text descriptions of the injuries. Sensitivity revision. Fiona Wood contributed to study design and critical
assessment suggests that they will not have biased the revision.
analysis which is the foundation of BISS. Other possible
sources of bias include inter-observer variability in TBSA
estimation and the selection bias of only those referred to a Conflict of interest statement
Trauma/Burn center [15]. Finally the BISS does not include
inhalation injury, a recognized mortality prediction indicator The authors declare no conflict of interest.
for burns. Inhalation injury was not considered in the model
development due to absence of consensus diagnostic criteria.
Despite the frequent index presentation of burns to a Acknowledgements
trauma center there has been limited research examining the
effectiveness of the ISS in predicting burn mortality [16,17]. The authors thank the staff of the Royal Perth Hospital Trauma
Others have reviewed ISS and burns in the context of other Service for access to the Royal Perth Hospital Trauma Registry
Trauma/Burn severity scores and have had smaller cohorts data and to Aaron Berghuber of the Fiona Wood Foundation,
[9,18]. Krob et al. reviewed 511 patients, reporting correlation Royal Perth Hospital for the linkage of the Burns Minimum
between ISS and outcomes but favored burn specific scores, Data Set and Royal Perth Hospital Trauma Registry data.
citing them as more accurate [18,19].
Prior to this study, the possibility of improving the ISS for
references
burns by incorporating TBSA and age had not been explored.
This is despite these being the two most established burn
mortality prediction variables [6,20,21]. A significant propor-
[1] Mock C, Peck M, Krug E, Haberal M. Confronting the global
tion of burns are managed at trauma centers where the ISS is
burden of burns: a WHO plan and a challenge. Burns
the most established mortality prediction tool. We propose 2009;35(August (5)):615–7.
that the BISS would potentially facilitate outcome comparison [2] Brusselaers N, Monstrey S, Vogelaers D, Hoste E, Blot S.
between burns managed in trauma centers and burn centers. Severe burn injury in Europe: a systematic review of the
To date this has not been possible. incidence, etiology, morbidity, and mortality. Crit Care
2010;14(5):R188.
[3] Miller SF, Bessey PQ, Schurr MJ, Browning SM, Jeng JC,
5. Future studies Caruso DM, et al. National Burn Repository 2005: a ten-year
review. J Burn Care Res 2006;27(July–August (4)):411–36.
[4] Gennarelli TA, Champion HR, Copes WS, Sacco WJ.
The BISS requires prospective validation, ideally in a multi- Comparison of mortality, morbidity, and severity of 59,713
center cohort and should be further validated in the combined head injured patients with 114,447 patients with
burn/trauma patient to confirm accurate concordance with extracranial injuries. J Trauma Injury Infect Crit Care
mortality. We also intend to pursue evaluations of ISS and BISS 1994;37(December (6)):962–8.
[5] Pacella SJ, Harkins D, Butz D, Kuzon WMJ, Taheri PA.
for other trauma outcomes such as length of stay and quality
Referral patterns and severity distribution of burn care:
of life.
implications for burn centers and surgical training. Ann
Plast Surg 2005;54(April (4)):412–9.
[6] Ryan CM, Schoenfeld DA, Thorpe WP, Sheridan RL, Cassem
6. Conclusion EH, Tompkins RG. Objective estimates of the probability of
death from burn injuries. N Engl J Med 1998;338(February
The ISS correlates with burn mortality but is a poor predictor (6)):362–6.
[7] American College of Surgeons Committee on Trauma. In:
of death with increasing severity of burn. This study
Rotondo M, Fildes J, editors. National Trauma Data Bank
developed a superior burn mortality prediction model (BISS) Annual Report. Chicago: American College of Surgoens;
combining ISS with age and TBSA. This model provides a 2012.
standardized mortality prediction for burn center patients and [8] Sampalis JS, Boukas S, Nikolis A, Lavoie A. Preventable
trauma center burn patients, potentially facilitating outcome death classification: interrater reliability and comparison
comparisons. with ISS-based survival probability estimates. Accid Anal
Prev 1995;27(April (2)):199–206.
[9] Baker SP, O’Neill B, Haddon WJ, Long WB. The injury
Contributions severity score: a method for describing patients with
multiple injuries and evaluating emergency care. J Trauma
Injury Infect Crit Care 1974;14(March (3)):187–96.
J. Tristan Cassidy contributed to study conception, literature [10] Pepe MS. The statistical evaluation of medical tests for
search, study design, paper composition, data analysis classification and prediction. Great Clarendon streed,
and interpretation. Michael Phillips contributed to study Oxford, UK: Oxford University Press; 2004.
[11] Akaike H. A new look at the statistical model identification. and relation to the Injury Severity Score. J Burn Care
IEEE Trans Automat Control 1974;19(December (6)):716–23. Rehabil 1989;10(November (6)):512–6.
[12] Hosmer DW, Lemeshow S. Applied logistic regression. [18] Krob MJ, D’Amico FJ, Ross DL. Do trauma scores accurately
Hoboken, New Jersey, USA: Wiley-Interscience; 2000. predict outcomes for patients with burns? J Burn Care
[13] Holm S. A simple sequentially rejective multiple test Rehabil 1991;12(November (6)):560–3.
procedure. Scand J Statist 1979;6:65–70. [19] Zimmerman TJ, Krizek TJ. Thermally induced dermal
[14] Simes RJ. An improved Bonferroni procedure for multiple injury: a review of pathophysiologic events and therapeutic
tests of significance. Biometrika 1986;73(3):751–4. intervention. J Burn Care Res 1984;5(3):193.
[15] Wachtel TL, Berry CC, Wachtel EE, Frank HA. The inter- [20] Muller MJ, Pegg SP, Rule MR. Determinants of death
rater reliability of estimating the size of burns from various following burn injury. Br J Surg 2002;88(November (4)):583–
burn area chart drawings. Burns 2000;26(March (2)):156–70. 7.
[16] The American College of Surgeons. National Trauma [21] Smith DL, Cairns BA, Ramadan F, Dalston JS, Fakhry SM,
Databank Annual Report 2010; 2011, July. Rutledge R, et al. Effect of inhalation injury, burn size, and
[17] Wong L, Grande CM, Munster AM. Burns and associated age on mortality: a study of 1447 consecutive burn patients.
nonthermal trauma: an analysis of management, outcome, J Trauma Injury Infect Crit Care 1994;37(October (4)):655–9.

Developing A Burn Injury Severity Score

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Developing A Burn Injury Severity Score

Uploaded by

Copyright:

Available Formats

JBUR-4186; No.

burns xxx (2013) xxx–xxx

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/burns

Developing a burn injury severity score (BISS):

J. Tristan Cassidy a, Michael Phillips b, Daniel Fatovich c, Janine Duke d,

article info abstract

2 burns xxx (2013) xxx–xxx

upon a coding lexicon originally developed by the Association

2nd degree; partial thickness <10% 1 1

2nd or 3rd degree; partial or full thickness 10–19% 2 4

2nd or 3rd degree; partial or full thickness 20–29% 3 9

2nd or 3rd degree; partial or full thickness 30–39% 4 16

2nd or 3rd degree; partial or full thickness 40–89% 5 25

burns xxx (2013) xxx–xxx 3

(1) An examination of the concordance of the ISS with

4 burns xxx (2013) xxx–xxx

Table 3 – Outline of burn deaths.

burns xxx (2013) xxx–xxx 5

6 burns xxx (2013) xxx–xxx

Total 25,844a 633 (2.4)

burns xxx (2013) xxx–xxx 7

8 burns xxx (2013) xxx–xxx

burns xxx (2013) xxx–xxx 9

You might also like