predicting death in burn patients death in burn pa'i'rznts - master ... multiple logistic...

71
Predicting Death in Burn Patients Manuel Gomez Ospina, M.D. A thesis submittd in conformity with the rquirements for the degree of Master's of Science, Graduate Department of Institute of Medical Sciences, University of Toronto. @Copyright by Manuel Gomez Ospina, ZOO 1

Upload: vuthuan

Post on 08-Jun-2018

230 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Predicting Death in Burn Patients

Manuel Gomez Ospina, M.D.

A thesis submittd in conformity with the rquirements for the degree of Master's of Science,

Graduate Department of Institute of Medical Sciences, University of Toronto.

@Copyright by Manuel Gomez Ospina, ZOO 1

Page 2: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

The author has granted a non- exclusive licence dowing the National Library of Canada to reproduce, loan, distn'bute or sel copies of this thesis in microform, paper or electronic fomats.

The author tetains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts fiom it may be printed or otheNYise reproduced without the author's pewissioa.

L'auteur a accordé une licence non exclusive permettant a la Bibliothbque nationale du Canada de reproduite, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/film, de reproduction sur papier ou sur format 6lectronique.

L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.

Page 3: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

PREDICTING DEATH IN BURN PA'i'RZNTS -

Master Science, 2001

Manuel Gomez Ospina, M.D.

Graduate Department of Institute of Medical Sciences, University of Toronto

ABSTRACT

Objective: To validate the FLAME (Fatality by Longevity, APACHE II score and

Measured Entent of bum) score for outcome prediction in bum patients.

Design: Multi-centre prospective study.

Setting: Five Canadian burn centres: Hamilton (ON), St. John (NB). Toronto (ON),

Vancouver (BC), and Windsor (ON).

Patiene 527 adult bum patients (1997-1998).

Outcorne: Hospi ta1 mortality . Resuits: The typical patient was a man (69%) with mean age of 44 years, mean body area

bwed of 17%, and mean hospital stay of 19 days. There were 76 deaths (14%). The

FLAME score was the simplest logistic mode1 with strong agreement between observed and

predicted deaths (ROC area = 0.96). Using 2x2 classification table with 0.5 predicted risk of

death, patients were comtly classified 94% of the time. Sensitivity: 68%, specificity: 9856,

and positive likeiihad ratio: 34 were found.

Conclusion: The FLAME score is a valid mortality predictive twl for bum patients.

Page 4: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

ACKNOWLEDGEMENTS = -- - A 2

This diesis is dedicated to Dr. Joel S. Fish, medical director of the Ross Tilley Burn

Centre program, at the Sunnybrodr & Women's College Health Sciences Centre, and

menber of my thesis comrnittee, who gave me the oppomuiity to increase my knowledge,

skills and experience in burn management, reconstructive surgery, and bum research. Joel's

continuous encouragement, wise advice and great enthusiasm, have been of paramount

importance for the completion of this research study. 1 am deeply indebted with my thesis

supervisor, Dr. Donald Redelmeier, whose expert advice, and invaluable support, had

inspird me to complete this thesis. 1 would also like to thank Dr. John Paul Szalai, member

of my thesis cornmittee, and Marko Katic, statistician, who gave me the extraordinary

statistical support and guidance for the analysis of this study data. 1 would also like to thank

Dr. David T. Wong, critical care specialist and anaesthesiologist, who gave me the

opportunity to leam, administer and evaluate the APACHE II scoring system in severely il1

critical care patients. Special thanks to Dr. Thomas Stewart, critical care specialist, who

stimulated me to conduct the initial retrospective analysis of outcome predictors in b u m

patients. 1 would like to express my sincere gratitude to the medical directors of this study

pdcipating bum centres: Dr. Janet Sproat (Hamilton, ON), Dr. H. Lalonde (St. John, NB),

Dr. James Boyle (Vancouver, BC), and Dr. B. Snowdon (Windsor, ON), for their invaluable

cooperation. 1 would like to thank the Work and Health Program of the Ontario Workplace

Safety & Insurance Board for awarding me the scholarship to complete this thesis. Lastly, 1

would like to thank my family, whose eternal love, pemanent support, unlimited patience

and generosity, have made possible the successfbl journey of my medical education, surgical

training, and professional activities.

iii

Page 5: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

TABLE OF CONTENTS ... . .

................................... .......................................*.. ABSTRACT .. .......................................................... ACKNOWLEDCEMENTS

LIST OF TABLES .................... .. .............................................. LIST OF FIGURES ......................................................................

LIST OF ABREVIATIONS .............................................................

1.0 INTRODUCTION ............................................................. ...................................................................... 1.1 Rationale

1.2 Study Question ............................................................. 1.3 Purpose ......................................................................

...................................................................... 2*0 BACKGROUND

2.1 Oveniew ..................................................................... 2.1.1 Economic Implications ....................................................

2.2 Outcome preâictive tools in Bum Patients .................................. ...................................................................... 2.3 FLAME Score

2.3.1 FLAME score derivation ....................................... ... .................................. 2.3.2 FLAME score preliminary validation

3.0 METHODS ...................................................................... 3.1 Study Design and Methodology ........................................... 3.2 Source of data and key variables ...........................................

3.2.1 Inclusion Criteria .................................................... 3.2.2 Exclusion Criteria ....................................................

3.3 Burn management clinical details ........................................... 3.4 Smoke Inhalation Injury diagnosis ........................................... 3.5 Statistical methods used for data analysis ..................................

4.1 Baseline Characteristics .................................................... 4.1.1 Survivors versus Non-swvivors ........................................... 4.1.2 Gender Differences ....................................................

4.2 FLAME score accuracy of outcome prediction ......................... 4.3 Contrasts between the multi-centre validation vs . derivation populations .

................................................... .......................... Tables .. .................................................. ................... Figures .. .. ...

ii

iii

vii

viii

Page 6: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

....................................................................... 5.0 DISCUSSION . . Sk-ELAME-- ~ ~ ~ ~ l t i e t r e ~rtlidrttion- ...................................

5.2 Cornparison of FLAME score to previous scores .......................... 5.2.1 Sirnilarities ..............................................................

5.2.1. 1 Age and percent TBSA ........................................... 5 . 2. 1.2 Gender ..................................... .. ......................

5.2.2 Differences ............................................................. 5.2.2.1 Full thickness bum ........................................... 5.2.2.2 Smoke inhalation injury ........................................... 5 .2. 2.3 Multiple ûrgan Failure ........................................... 5.2.2.4 Other multivariable scores ..................................

6.0 CONCLUSIONS ...................................................................... 49

6.1 Main conclusion ............................................................. 49 ...................................................................... 6.2 Implications 49

7.0 LIMITATIONS ...................................................................... 50

7.1 Few Participating Bum Centres ........................................... 50 7.2Singlecountrystudy ............................................................. 50 7.3Qualityoflife ...................................................................... 51 7.4 Ethicai considerations ............................................................. 52 7.5 Adult population ........................................................... 52 7.6 Self-selection of participating bum centers .................................. 52

8.0 FUTURE RESEARCH ............................................................. 53 8.1 Tests in other populations ................................................... 53

.................................................... 8.2 Explore ethical implications 53 8.3 Explore economic issuess .................................................... 53

REFERENCES

Page 7: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

+ ..

Table 1

Table 2

Table 3

Table 4

Table 5

Table 6

Table 7

Table 8

Table 9

Table 10

Table 11

Table 12

Table 13

Table 14

Table 15

LIST OF TABLES . . . . .

................................ APACHE II score sheet ... ................. 23

Multiple logistic regression anaiysis O 1990- 1995 (n = 447) ............ 24

Age. APACHE II. and TBSA codes .................................. 25

Two by two diagnostic table for the FLAME score (derivation population.

1990- 1995) (n = 474) with decision criteria of 0.5.0.7. and 0.9 ....... 26

....... FLAME derivation versus preliminary validation populations 27

Two by two diagnostic table for the FLAME Score: derivation vs . preliminary validation populations with 0.5 decision criteria ....... 28

Participating Canadian burn centres population ( 1997- 1 998) ..... 29

Survivors versus non-survivors ........................................... 30

.................................................... Differences by gender 31

Diffaences in non-survivors by gender .................................. 32

Multiple Logistic Regression results for the mode1 that includes FLAME + Centres + Sex .................................................... 33

Multiple logistic regession models (n = 527) ......................... 34

Two by two diagnostic table for the FLAME score (multi-centre validation. 1997- 1998) (n = 527) with decision criteria of 0.5,0.7,

...................................................................... and 0.9 35

FLAME derivation vs . FLAME multi-centre validation populations 36

Two by two diagnostic table for FLAME score: multi-centre

vaiidation vs . denvation populations with 0.5 decision criteria ....... 37

Page 8: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

LIST OF FIGURES ........A - . ... .

Figure 1 Probability of death by FLAME score and gender ............... 38

Figure 2 FLAME score ROC curve - Derivation population (n = 474) ......... 39

Figure 3 FLAME swre ROC cuve- Preliminary validation population (n= 165) 40

Figure 4 FLAME score ROC c w e - Multi-centre validation population

(n=527) ..................................................................... 41

Figure 5 ROC curves: Derivation vs . Multi-centre validation ............... 42

Figure 6 Adult burn admissions . 1999 . Cross Canada survey ............... 43

vii

Page 9: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

LIST OF ABREVIATIONS

APACHE Acute physiology and chronic health evaluation score

FLAME Fatality by Longevity. APACHE II score and Measured Extent of bum

ROC Receiver operating characteristics curve

RTBC Ross Tilley Burn Centre

TBSA Percent total body surface area bumed

Page 10: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

CHAPTER 1

INTRODUCTION

The p u p s e of this chapter is to:

1. discuss the rationale of the study

2. introduce the stutiy question

3. describe the purpose of the stu&

4. in~odicce the stu@ hypothesis

1.1 Rationale

Prediction of outcome in burn patients following admission to a bum centre may be

useful to determine the patient requirements of nursing care, diagnostic modalities,

medical treatment, surgical interventions, and rehabilitation therapy. It may facilitate

communication with the patient, the family, and other health care professionais. Accurate

prdictions may also be usehl in planning transfers to major burn centres and in scarce

resowce allocation decisions.

1.2 Sîudy Question

1s the FLAME (Fatality by Longevity, APACHE II score and Measured Extent of

bum) score an accurate predictive tool of death in Canadian bum patients? Answering

this question may contribute to better allocation of human and technological resources,

triage and management decisions, evaluation of new diagnostic and treatment modalities,

and cornparison of burn populations studies.

Page 11: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

The aim of Oiis study was to vdidate the FLAME score for outcome prediction in

adult burn patients from different Canadian b u m centres.

1.4 Hypothesis

The FLAME score is a valid outcome predictive tool for burn patients.

Page 12: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

BACKGROUND

The pupose of d i s chapter is to:

1. provide an oventiew of the burden of illness associated with burn injuries

2. discuss dl fferent outcome predictive tools previousiy developed

3. describe the derivation und preliminay validation of the FLAME score

2.1 Ovewiew

Burns are devastating injuries that can produce permanent disfigurement, physical

dysfûnction, psychological morbidity, and death. Each year, approximately 200,000

Canadians receive medical care for burn injuries,' more than five thousand people (18.9

per 100,000 population) are hospitalized for an average of 13 days, with a mortality rate

of 4.1%: Early prediction of outcome (i.e., survival or mortality) may help triage

effectively, and to implement early medical and surgical interventions efficiently. In tum,

better care could reduce hospital complications, shorten length of stay, improve quality of

life, and enhance survival.

2.1.1 Economic Implications

The estimated cost of an average bum unit bed in Canada is about $1,800 per day,

although some days may cost much more depending on the intensity of care.

Expenditures of about S 1 l7,OOO,Oûû are incurred for the 5,000 patients hospitalized every

year in Canada? The economical burden of subsequent recovery is also substantial. Burn

Page 13: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

--- -= -- A suyivoe often require long-tem rehabijitation to retum to their pre-bum level of

activities, quality of life, and independence. In addition, repeated reconstructive surgeries

may be required to release bum scar contractures, remove hypertrophic scars, improve

physical mobility, enhance external appearance, improve self-image and promote

reintegration to society.

2.2 Outcome predictive tools in Bum Patients

More than half a century cgo Bull and squire3 developed a prognostic index for

themurlly injured patients that reflected the importance of age and percent total body

surface m a burned (TBSA) in mortality. Using probit analysis, Bull and Squire

cdculated the approximate mortality probability for different age groups (0-1 4 years, 1 5-

44 years, 45-64 years, 65 and more years) and derived a grid of approximate mortality

probabilities for various combinations of age and TBSA.' Using the Bull and Squire grid,

a 62 year old man with 13% TBSA would have a 40% probability of death. This index

allowed comparative assessments of different groups of themally injured patients.

The Bull and Squire index was followed tifteen years later by the first predictive

equation including more patient variables: age, percent area of third degree burn

(determined at the end of hospitalization), and post-bum complications (cerebral,

pulmonary, or ~hock).''~ Using multiple regression techniques, regression coefficients

were determined for each variable, and with probit analysis, probit equations were

fomiulated for each age group and percent of third degree bum. Thus, for a 62 year old

man with 13% third degree burn without post-bu. complications, the calculation is:

Page 14: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

3.310 + (62 - x 0.45288) + (13 x 0.786) + (O x 0.236) + (O x 0.178) + (O x 0.149) = 41.6.

Therefore, this person would have a 42% probability of death.

The Baux's rule (quoted in Stem and waisbren6) a simple rule of thurnb that adds the

patient's age to the TBSA has been used in burn treatment for many years. This rule

assurned ihat values of more than 75 indicate poor prognosis with greater than 50%

probability of death. Using the Baux's nale in a 62 years old man with 13% TBSA, the

calculation is: 62 + 13 = 75. Therefore, this person would have less than 50% probability

of death.

A modified Baux d e 6 excludes patients less than 20 years of age (arguing that

mortality in younger patients does not decrease with lower age), and assumes that a

patient has a greater than 50% probability of death if the sum of age and TBSA exceeds

95. Using the modified Baux's nile in a 62 yem old man with 13% TBSA, the

calculation is: 62 + 13 = 75. Therefore, this person would have less thsn 50% probability

of death.

Many investigators have f.urther analyzed mortality predictors in thermally injured

patients combining age and TBSA with the following variables: gender,'-" percent full

thickness buni45$.1&22 and presence of smoke inhalation injury. ' ' J ~ * ~ ~ ' " ~ ~ ~ ~ ~ ~ 0th er

variables have been included in mortality predictive models: body weight,I0 prior

bmncopulmonary disease, a b m a l P,O,, and airway ede~na,'~ length of hospital stay,"

admission white blood count, admission senim osrnolality, involvement of inflammable

liquid, pre-existing mental disorders, circulatory disease and digestive diseme:' alcohol

abuse, nicotine abuse, preexisting newlogical and cardiovascular diseases," being pyo-

Page 15: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

prpne (Le., prone to infection by Pseudornonas pyocyanea)~ institution: absolute

monocyte count, absolute lymphocyte count, maximum daily temperature, and blood urea

nitmgen (BUN)." wound closure index," senun amyloid A protein:' and multiple organ

fail~re.'~"%ese previous authors slightly improved mortality prediction including more

patient variables.

in a previous Canadian studg6 on 1705 patients, Tredget showed that smoke

inhalation injury increased patient mortality compared with patients without inhalation

injury (34.7% vs. 1.7%, p <0.01), and increased length of hospitalization (74.4 days vs.

23.7 days) in survivors. But their patients with smoke inhalation injury were older (30

years vs. 25 years, p>0.05), and had significantly larger bums compared to patients

without smoke inhalation injury (39.7% vs. 12.2%, p ~ 0 . 0 1). After adjusting for age and

TBSA, the authon concluded that smoke inhalation injury increases bum mortality

independent of patient age and burn size.

Ryan and col leag~es~~ in a retrospective review of 1665 patients (54.7% children and

45.3% adults) found that age greater than 60 years, more than 40% TBSA, and inhalation

injury, were sipificant risk factors for mortality. Their mortaiity formula predicts patient

moaelity at 0.3% with no risk factors, 3% with one risk factor, 33% with two nsk factors,

and 90% with three factors. This nile is applicable to al1 patients younger than 90 years of

age. This nile was prospectively validated in other 530 burn patients from the same two

institutions of the derivation population. The weakness of this rule is that it only Oves

four levels of mortality risk instead of individual estimates w i t b a continuous range2'

which would allow the cdculation of the discriminant power of the formula (area under

Page 16: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

ROC ~ u r v e ) ~ ' ~ ~ and its calibration CHosmer-Lemeshow goodness-of-fit test)39, which - - --

were not mentioneâ in the study.

More recently, 0'lCeefeM using data from 4927 bum patients, detemined that the best

mode1 for estitnating burn mortality included: percent TBSA, percent full-thiclmess bum,

age categories of <30 years, 30-59 years, > to 60 years, and gender. In addition, O'Keefe

found that the risk of death was two-fold higher in wornen aged 30 to 59 years than men

with the same age.

Over the past five decades, the outcome of burn patients has improved as a result of

better understanding and management of burn shock,"~'~ use of more effective topical

antimicrobials," better antibioti~s,~ organization of regional burn unit^,"^ earlier

excision!'" and alternative measures for b m wounds c l o s ~ r e . ~ ~ For exarnple, in 1949,

half of young bumed adults died with 43% TBSA.' Forty five years later, half of young

bumed adults required more than 80% TBSA to die.'"

Many scoring systems have become popular in the intensive care setting Iike the

acute physiology and chronic health evaluation II score (APACHE The APACHE II

score has three components: acute physiology score (calculated h m the worst value of

12 physiologicd variabtes during first 24 hours afier admission, maximum 60 points),

chronic health score (prernorbid major organ dysfiinction, maximum 5 points), and age

score (maximum 6 points) (Table 1). The range of APACHE II score is h m O to 71

points. There is a direct relationship between the APACHE II score and hospital death

rate.&*" The APACHE II score has been validated worldwide in adult cntically il1

patients as a predictor of group mortality based on the first day of ICU stay."'-'*

Page 17: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

The developers of the APACHE scoring system excluded burn patients." Despite this - - -A .-. - - *

initial exclusion, the APACHE II score has sometimes been measured in thennally

injured patients. CannonJ9 in a prospective study of 3 1 adult burn patients, did not find

any correlation between plasma circulating interleukine4 and tumor necrosis factor*

with the APACHE II score h m day 1 to 15 following burn injury. Brown6" found a

significant correlation (p<0.001; 8=0.20) between the APACHE II score and physiologie

stress measured by the resting energy expenditure, in a prospective study of 70 adult

critically il1 patients, which included 11 burn patients. SameJ5 found a better outcome

correlation with the Thermal Injury ûrgan Failure Score (?-=0.66) than with the

APACHE 11 score (3~0.15) during the first week of treetment, in 3 prospective study of

83 adult bum patients. Viscardi6' found a good comlation between the APACHE II score

and mortality (non-survivon APACHE 11 = 1 9 vs. survivors APACHE II = 1 5) in a case

controlled study of 30 severely bumed patients with large bums (50% body surface area),

h m which 15 required partial limb amputations.

2.3 FLAME Score

2.3.1 F W E score derivation

Our group first conducted a retrospective cohort study of 474 adult burn patients

discharged h m the Ross Tilley Burn Centre (RTBC), the regional refmal centre for

adult burn patients h m the metropolitan Toronto area and Northern Ontario between

1990 and 1995. We reviewed the significant independent predictor factors of hospital

mortality using a stepwise multiple logistic regression analysis, with hospital death as the

Page 18: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

dependent variable. Incrrased age, TBSA, -- -- - - day 1 APACHE II score, and fernale gender

were found to be independent predictors of increased mortality (Table 2)."

Based on the results of the multiple logistic tegression analysis, the signifiant

independent variables were multiplied by the coefficients obtained, and the resulting sum

across al1 the variables produced a value called l~git. '~ A togit is the natural logarithrn of

the ratio of two probabilities (e.g., dead vs. ali~e).'~

Therefore, logit = y = bo + b,x, + b,x,. . . . . .bix, where y is the dependent variable

(hospital mortality), b, is a constant, b, through bi are coefficients, and x, through xi are

the individual predictors.

Then, we calculateci the probability of death for each patient using the following

eq~at ion~~: Pr (hospital mortality) = elogit /(1 + elwi?

The following equation was established to identify patients with more than 50%

probability of death: y = 8.7801 - (0.4352 x AGE CODE) - (0.8169 x TBSA CODE) -

(0.8219 x APACHE II CODE) + (1.0837 x SEX) where SEX = 0,l (female, male

respectively), and the codes for age, percent total body surface area bumed, and day 1

APACHE II score are found in Table 3.

To simpIi@ the equation, we rounded each factor to the nearest 0.1 and muttiplied it

by 2.5 obtaining the FLAME score:

(AGE CODE) + 2 (APACEE iI CODE) + 2 (TBSA CODE)

Based on this equation, the minimum score anyone could achieve is 5, and the

maximum is 49. The cut-off for 50% probability of death for wornen was 2 1 and for men

24. With these cut-offs this equation gives a sensitivity of 7 1%, a specificity of 98%, a

Page 19: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

- -La - positive predictive value of 8496, a negative predictive value of 96%, correct

classification of 95%, and a positive likelihood ratio" of 35.5 (Table 4).

For a practical clinical application, and based on our study results (Le., predicted vs.

observeci mortality), we classified the mortality risk for thennally injured patients

according to the following severity levels:

MortaUty Risk F U M E Score

Minor ( 0 - 25%) 5 - 18

Moderate (26 - 50%) 19 - 21

H m ( 251% ) 22 - 49

Exampie

In our b m population, a 62 year old man with a TBSA of 13% and an APACHE II

score of 7, would have a FLAME score of (5)+2(2)+2(2)= 1 3. As this FLAME score falls

between 5-1 8, he would have a minor risk of mortality (0925%).

Mortality curves were created by finding the average probability of mortality at each

five point interval for women and men (Figure 1). The curves show that scores ranging

between 15 to 25 for women and 20 to 30 for men place patients on the steepest part of

the mortality curve. Women died with lower score (10-25) than men (26-28).

Using the receiver operating characteristic (ROC) cuve analysis,'"* we found an

area under the ROC curve f standard error of 0.98 f 0.03 for the new predictive

equation?(Figure 2)

Page 20: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

2.3.2 F w score prelhinav validation - -

The reproducibility of the FLAME score was assessed in a prospective data h m 165

bum patients admitted to the RTBC during the year 1997. This validation population had

similar characteristics than those found in the derivation population (Table 5). The only

differences were that the validation population had significantly higher proportion of

non-survivors (22 vs. 12, p=0.003), fewer bums due to other etiologies different than

flame and scald bums (e.g., electrical, contact, tar, chemical, lightning, and radiation,

17% vs. 22%, p = 0.047), larger extension of full thickness bums (1 1% vs. 8%, p =

0.044), required fewer days of mechanical ventilation (4 days vs. 18 days, p < 0.00 l), and

required more operations (1.2 vs. 1, p = 0.014), than the denvation population. There was

a good correlation between predicted and observed outcornes, with an area under ROC

c w e f standard error of 0.97 f 0.04 (Figure 3), sensitivity of 61 %, specificity of 99%,

correct classification of 9 1%, and a positive likelihood ratio of 6 1 (Table 6).

This preliminary validation of the FLAME score in RTBC patients motivated us to

ascertain generalizability of the FLAME score by conducting an extemal validation of the

FLAME score in adult burn patients with different characteristics and treated in different

b m centres with different characteristics than those of the RTBC, hence, the purpose of

this shdy.

Page 21: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

CH-Wl'ER 3

METHODS

The pupose of dis chapter is to:

1. provide an overview of the snufy design and rnethodologv

2. describe the source of data and the key variables

3. discuss the stutistical methods used for data analysis

3.1 Study Design and Meihodology

This was a prospective multi-centre cohort study conducted to validate the predictive

accuracy of the FLAME score in burn patients.

3.2 Source of data and key variables

After approval by the Ethics Review Board of the Wellesley-Central site of the St.

Michael's Hospital in Toronto, an invitation letter was sent to the medical directors of the

other 26 Canadian burn centres.' Participation consisted in data collection of

demopphic variables (e.g., age, gender), bum charactenstics (e.g., date of bum,

etiology, percent TBSA burn defined as the sum of partial and full thickness bum,

percent fulI taickness burn, smoke inhalation injury), outcome variables (e.g., days of

mechanical ventilation, length of hospitai stay, hospital mortality), and data for the

calculetion of the APACHE II score (e.g., 12 physiologie parameters in the first 24 hours

pst-admission, pre-burn medical condition, anâ admission age), of adult patients

admitted behueen January 1997 and December 1998. Data was received h m five

Page 22: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

centres: Hamil ton (Ontario), St. John (New Brunswick), Toronto (Ontario), Vancouver ---. .... 2 - . -- - - - . - - - .- - - - - - -

(British Columbia), and Windsor (Ontario) (Table 7).

Many of the non-participating bum centre directors claimed that they did not have a

person that wuld collect and send the required data to our burn centre. Other medical

directors explained that they admitted few patients per year who were eligible for this

study. ûniy one medical director declined to participate for personal reasons.

3.2.1 Inclusion cri teria

Al1 adult patients with acute burn injuries discharged h m the participating Canadian

burn centers during the study period (1997-1998) were includeci in this study. Patients

with "do not resuscitate" orders who had comfort measwes treatment were also included

in the analysis.

3.2.2 Exclusion criteria

Patients discharged during the snidy pend with incomplete data were excluded h m

the study. In addition, patients discharged with other diagnostics different than acute

bums (e.g., toxic epidermaI necrolysis, necrotizing fasciitis), or who sere admitted for

reconstmctive pst-bum surgery were also excluded h m the study.

3.3 Burn management clinical details

Al1 patients were treated according to the following Canadian standard of burn

centres care, with initial fluid resuscitation using intravenous Ringer's lactate solution,

Page 23: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

acco-@bg to a modified Parklagd f o ~ p l a @c x k g x %TBSA) with additional fluids

given as necesS81y to obtain a minimum urine output of 0.5 cc's per kg per hour." Eiîrly

nutrition (4 day pst-burn) was suppliai through oral or enteral feeding. Dressing of

buni wounds were performed two or thtee times a day using topical antimicrobials: silver

sulphadiazine (SSD), rnafenide (a-amino-ptoluene sulfonamide monoacetate) or

polysporin @olymyxin B sulfate & bacitracin). Early surgical excision of bum eschar and

early coverage of exciseâ bum wounds with autopfis was perfonned between two to

five days pst-bum injury? Biobranea, (a synthetic by-layer with a porcine collagen

based substitue) was used as a temporary wound coverage in some cases. Donor sites for

skin grafis were âresseà with scarlet redw (oil impregnated fine mesh). In addition,

medical spezialty consultations (e.g., cardiology, endocrinology, psychiaüy, etc.) and

social service assistance were provideci when needed.

3.4 Smoke inhalation injury diagnosis

For the purpose of this multi-centre study, we considered a positive diagnosis of

srnoke inhalation injury only with the clinical diagnosis (e.g., bumed in a closed space,

facial bums, sooth in mouth), because not al1 Canadian bum centres perfom

bronchoscopy to confimi the diagnosis of smoke inhalation injury within the first 24

hours &er bum injury.

Page 24: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

3.~tatisticai methods used for data analysis

Descriptive statistics were used to determine the means, standard deviations and

ranges h m al1 the variables. Cornparisons betwen survivors and non-survivors were

made using the un-paired Student's t-test for continuous variables (e.g., age, percent

TBSA, length of stay in hospital, days of mechanical ventilation) and the chi-square test

of independence for binary variables (e.g., presence of smoke inhalation, mortality). A

two tailed ~ ~ 0 . 0 5 was considered statistically significant.

The probability of death was calculated af€er the first day of admission for each

patient, using the FLAME score", and then compared with the observed mortality. The

predictive power of the FLAME score was evaluated by multiple logistic regression

anaiysis, two by two diagnostic table, the area under the ROC c u ~ e , " ~ ' the Hosmer-

Lemeshow goodnesssf-fit statistics," -2 log likelihood ratio,M and the Brier score?

A h transfortning categorical variables to dummy variables (e.g., female = O, male =

1), hieratchical multiple logistic regression models were evaluated to determine which

risk factor, in addition to the FLAME score, improve the predictive power. The initial

model was the FLAME score alone. Then, we added other possible predictive variables:

bum centres, sex, etiology, inhalation injury, days of mechanical ventilation, days

intemal from burn injury to admission, and the last mode) had al1 variables. For each

model we detemined the Schwartz d rite non:^ the Hosmer-Lerneshow goodness-of-fit

statistics," -2 log likelihood ratio," and the area under ROC,"^'

The Schwartz CntezionM is used to test whether the dependent variable (e.g.,

mortaiity) is significant, based on a chi-squared distribution adjusting for the number of

Page 25: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

exp&natory variable and the n e b g of OB-ations -@ in the model. This statistics is - -

usefùl for comparing different models; lower values of the Schwartz Cntenon indicate a

better fitting model."

To describe how faithfully the predicted and observed results compare overall using

the FLAME score, the Hosmer-Lemeshow goodness-of-fit statistics was ~sed."~ This

statistics is obtained by calculating the Pearson chi-square statistics from the 2 x g table

of observed and expected frequencies, where g = 10 is the number of groups fomed by

deciles of risk of death. A high p value means a good model because the hypothesis that

the data fit the specified model was not rejected. The higher the p value the better the fit

of the model.

Receiver operating characteristic (ROC) c w e was constnicted for the FLAME score

h m the patients predicted and observed hospital outcornes. A plot of ûue positive rate

(sensitivity) against false positive rate (1-specificity) was made and the area under the

ROC was derived. The area under the ROC is a measure of the overall discnminatory

power of the prognostic variable (0.5 = random discrimination, and 1.0 = perféct

di~crirnination)."~~ Most specifically, if one would take random pairs of patients where

one did survive, and other did not, then the predicted suMval score would correctly

identify the non-surviving member of the pair with the probability of the area under the

ROC curve. For exarnple, 0.5 reflects the chance level of the above identification (e.g.,

5060). A significant area under ROC curve describes an increase of the probability

above 0.5.

Page 26: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Two by two diagnostic table we? co~truct@ to-compire the accuracy of prediction - - - ---

of outcome, using the FLAME score with a decision criteria of 0.5 as cut-off. At this cut-

off, patients with a calculated probability of death greater than 0.5 were predicted to die,

while those with a calculated probability of death equal or less than 0.5 were predicted to

survive. We then detennined the sensitivity, specificity, positive predictive value,

negative predictive value, and correct classification of the FLAME score. In addition, we

calculated the positive likelihood ratio to determine the odds that a given predicted risk of

death by the FLAME score would be expected with the observed death? It is calculated

dividing sensitivity by 1 00-speci ficity . The closest the resul t of the positive likeli hood

ratio is to 4 the better the performance of the test.

The -2 log likelihood ratio (also called devianceb8) measures unexplained vanability

in the data and thus lower values indicate a better fit." We will use a chi-square test to

determine the difference between the -2 Log likelihood ratio of the FLAME logistic

mode1 alone with each of the other models separately .

The Brier score was useci to quanti@ the accuracy of the FLAME score by comparing

predicted with actual outcornes. A Brier score of O indicates perfect agreement between

predicted and actual outcome and a Brier score of 1 indicates perfect disagreement." A

Brier score of 0.25 could be possible when we assign a probability of 50% to every

patient ."

Ail analyses were performed using the SAS for windows version 6.12."

Page 27: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

CHAPTER 4

RESULTS

The purpose of this chapter is to:

1. compare baseline characteristics of survivors and non-survivors

2. assess the accuracy of outcome prediction of the FLAME score

3. contrast the mufti-centre validation resuh with those of the derivation

We received data from 544 patients discharged from the fwe partici pating Canadian

burn centres. We excluded 14 patients because they were children (4 5 years), 2 patients

becawe had other etiology different than bums (e.g., Toxic Epidennal Nemolysis,

Stevens Johnson Syndrome), and 1 patient because had incomplete data. Therefore, for

the analyzes of this study we included data hm 527 patients.

4.1 Basetine Characteristics

The study population had a mean age of 44 years (range t 5 years to 93 years), with a

majority of men (69%). The most comon bum etiology was flame (62%), followed by

scald ( 1 6%), elecûical ( 1 O%), contact (5%), tar (4%), chernical (3%), lightning (0.2%),

and radiation (0.2%). The mean percent TBSA was 17% (range 0% to 100%). Smoke

inhalation injury was premt in 23% of the patients (78.5% confinned by bronchoscopy),

and 29% of the patients required mechanical ventilation. The mean length of hospital stay

was 19 days (range 1 day to 470 days), and there were 76 deaths ( 14%) (Table 8).

Page 28: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

4.1. I Survivors versus Non-suwivors

Flame and scald bums accounted for almost al1 of the deaths (75 of 76,99%). Men

were less likely to die (46 of 364, 13%) than women (30 of 163, 18%), but this trend was

not statistically significant. Non-survivors were older (58 years vs. 4 1 years, p c0.00 l),

had more fiame burns (82% vs. 59%, p ~0.00 l), had larger burns (TBSA = 44% vs. 12%,

p c0.001), had deeper burns (full thickness burn = 34% vs. 3%, p <0.001), had more

incidence of inhalation injury (54% vs. 18%, p <O.OOl), were more likely to require

mechanical ventilation (84% vs. 20%, p ~0.001) during more days (8 days vs. 2 days, p =

0.004), and were more severely il1 (APACHE 11: 19 vs. 4, p <0.001; FLAME: 23 vs. 9, p

<0.00 1 ), than s u ~ v o r s (Table 8).

4.1.2 Gender Direrences

Comparîng men with women (Table 9), men accounted for 69% of the total

population, and had significantly more bums due to other etiologies different than flame

and scald bums (e.g., electrical, contact, tar, chemical, lightning, and radiation, 27% vs.

12%, ~ ~ 0 . û û I ) . However, women were significantly older than men (49 years vs. 41

years, pO.OOl), and had significantly more scald bums than men (23% vs. 13%.

p<o.Oo 1).

Analyzing gender differences in non-survivors, women had proportionally more

fatality than men (1 8% vs. 13%, p=0.082, Table 9), and died with significantly smaller

bums than men (TBSA = 35% vs. 5096, p=O.OlS, Table 10). However, men were more

Page 29: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

likely to require mechanical ventilation than women (93% vs. 70%, p=0.006), and were - .s - - -- -

more severely il1 on admission (FLAME = 25 vs. 2 1, p=0.0 16, Table 1 O).

4.2 FLAME score accuracy of outcome prediction

FLAME has been shown to have a strong predictive relationship to death (Odds Ratio

= 1.58, p = 0.0001 Table 1 1, area under ROC = 0.96 Table 12) in a well fitting model

(Schwartz Cnterion = 183, Hosmer-Lemeshow goodness-of-fit p = 0.49 Table 12).

Hierarchical logistic regression analyses indicated that adding centres to the FLAME

increased the predicting model slightly (area under ROC = 0.97, p = 0.004 Table 12).

Although, the combined group of burn centres has a positive influence in the model, none

of the individual centres alone were significantly related to hospital mortality in the

model (Table Il). The big standard e m r for Windsor (675 Table 11) and St. John's

(495.7 Table I 1) reflect the absence of reported deaths in these two bum centres (Table

7). Sex is the other variable with strong relationship to death (p = 0.0239 Table 1 1, Odds

Ratio = 0.36 Table 1 I), being male is protective.

Inspection of Table 12 suggests that although adding more variables to the FLAME

generally improves the model in ternis of p values, it has seemingly slight effect on the

area under ROC curve (fiom 0.96 to 0.98 Table 12). With the exception of the Hosmer-

Lemeshow goodness-of-fit p value for FLAME + Centres (p = 0.57 Table 12) and

FLAME + Centres + Sex (p = 0.54 Table 12), it may indicates that the FLAME alone is

less over fit.

Page 30: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

.- Although the last model, which - - includes al1 variables, - - had a better area under ROC

curve than the FLAME model (0.98 vs. 0.96) it was the worst fitting model (Schwartz

Criterion = 247) and it is not a practical clinical model.

We constructed ROC curves for three of the participating bum centers, which

reported deaths. Vancouver had the best area under ROC curve I standard e m r (0.982 f

0.0 14) followed by Toronto (0.969 f 0.0 10) and Hamilton (0.843 t 0.1 14).

Using a two by two diagnostic table with a predicted tisk of death of 0.5, patients

were correctly classified 94% of the time. The sensitivity was 68%, the specificity 98%,

and the positive likelihood ratio 34. (Table 1 3).

We found a very low Bner score for the observed mortality (0.048) compared to the

predicted mortdity (0.046) of the FLAME score. In addition, there was no significant

diffeience between the FLAME predicted mortality and the obsewed mortality @=-0.37).

These results support the accuracy of the FLAME score for mortality prediction in this

multi-centre bum population, because it failed to reject the hypothesis that there is good

apement between the FLAME score predicted mortality and the observed mortality.

4.3 Contrasts Between the Multi-centre Validation vs. Derivation Populations

Comparing the two populations, there were no signifiant differences in the

proportion of survivors, non-survivors, gender, mean age, incidence of flame burn, mean

depth of the burn, need for mechanical ventilation, and mean length of stay (Table 14). In

the derivation population, there were significantly more incidence of scald burns (26%

vs. 16%, p<O.ûûl), patients had larger burns (TBSA = 19% vs. 17%, p=0.035), patients

Page 31: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

--A=L - A - - reguircd more days of mechanical ventilation Cl 8 days vs. 3 days, p=0.001), were more

severely il1 (APACHE II = 8 vs. 6, p<0.001; FLAME = 13 vs. 11, p=0.005), and there

were 58 (12%) deaths. The validation population had significantly more bums due to

other etiologies different than flame and scald bums (e.g., electncal, contact, tar,

chernical, lightning, and radiation, 22% vs. I5%, p=0.006), had more incidence of

inhalation injury (23% vs. 12%, p<O.OOl), required more operations ( 1 vs. 0.7, p<0.001),

and there were 76 (14%) deaths.

The FLAME score had similar accuracy of outcome prediction in both populations,

measured by the area under ROC curve f standard enor (0.96 * 0.9 1 vs. 0.98 f 0.03,

Figue 5), similar correct classification (94% vs. 95%), similar sensitivity (68% vs. 7 1 %),

equal specificity (98% vs. 98%), and equal positive likelihood ratio (34 vs.34) (Table 15).

Page 32: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 33: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 2 Multiple Logistic Regremion Analysis - 1990-1995 (n474) L -- -- - . -- - --A-p -- -

Parrimeter Od& Ratio 95% Confidence p Value Intervals

Age 1.54 1.18 - 2.01 0.0014 TBSA 2.26 1.71 - 3.00 0.000 1 Women 2.96 1 .O2 - 8.53 0.045 1 Day 1 APACHE II 2.27 1.61 - 3.21 0.000 1

TBSA = percent total body surface area bumed APACHE II = acute physiologie and chronic health evaluation II score

Page 34: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 3 Age, APACHE II and T-A-codes

C& Age(yean) APACHE Il TBSA (./a)

TBSA = percent total body sutface area bumed APACHE II = acute physiologie and chronic health evaluation II score Code = equal intervals in ascending order

Page 35: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 4 Two by hvo dhgnoetie tabk for tbe FLAME Score (derivition popdation, 1990-1!M) (n = 474) witb decision criteria of 0.5,0.7, d 0.9

-

Preàicted rbk 0.5 Predicted rbk 0.7 Predicted risk 0.9

Observed Observed Observed

Preàicted Dead Alive Predicted Dead Alive Predicted Dead Alive

Sensitivity Specificity Correct Classification False positive False negative Positive predictive value Negative predicted value Positive likelihood ratio

1

FLAME = Fatality obtained by Longevity, APACHE II, and Measwed Extent of bum

Page 36: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

b b b t r n 0 - ~ ~ ~ $ $ 2 ! 8 ~ ~ ~ ~ 8 ~ y q ~ o o o o ' o o o o o o o o

Page 37: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 38: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

T m 7 Pa-wpating Q ~ a d i a n bu-rn c e a e po~ulation (1997-1998)

Burn Centre Patients Deaths No. (94) No. (%)

Hamilton (Ontario) 74 ( 14.0) 6 ( 7.9)

St. John (New Brunswick) 24 ( 4.6) O ( 0.0)

Toronto (Ontario) 330 ( 62.6) 65 ( 85.5)

Vancouver (British Columbia) 86 ( 16.3) 5 ( 6.6)

Windsor (Ontario) 13 ( 2.5) 0 ( 0.0)

Total 527 (1 00.0) 76 (100.0)

Page 39: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

{

Table 8 Survivors versus Noirauwivors i. c

VARIABLE ALL PATIENTS SURVWORS NONSURVWORS p Vdue * 1

Patients [n (%)] Gender Men [n (%)] Women [n (%)]

Age (Mean years * SD) Etiology of thermal injury

Flarne [n (%)] Scald [n (%)] ûther [n (%)]

TBSA (Mean % * SD) FïB (Mean % * SD) InhdaoOn injury [n (%)] N d for mechanicl ventilation [n (O41 Mechanical ventilation (Mean days * SD) Numbcr of operations (Mean * SD) LOS (Mean days + SD) Diy 1 APACHE II (Mean fi SD) FLAME (Mean * SD) p Value * = cornparison between survivors and non-survivors (chi-square test or t-test) Other = elecüical, contact, tar, chemical, lightning, and radiation TBSA = percent total body surface area burned W B = percent full thickness body surface area bumed LOS = length of hospital stay in days APACHE II = acute physiology and chronic health evaluation II score FLAME = Fatality obtained by Longevity, APACHE II, and Measured Extent of bum

Page 40: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 9 DifEerences by Gender

VARIABLE

L

t

ALL PATIENTS MEN WOMEN p Value * 1.

Patients [n (%)] Survivon [n (%)] Non-rurvivorr [n (%)] Age (Mean years SD) Etiology of thermal injuy

Flame [n (%)] Scald [n (%)] Other [n (%)]

TBSA (Mean % * SD) m B (Mean % * SD) inhalation hjury [n (%)] Need for mechmicil ventüation [n (%)] Mechadcil ventilation (Mean days + SD) Nuaber of operations (Mean SD) LOS (Mean days A SD) Diy 1 APACHE II (Mean * SD) FLAME (Mean * SD) p Value * = cornparison between men and women (chi-square test or t-test) Ocber = electrical, contact, tar, chernical, lightning, and radiation TBSA = percent total body surface area bumed FTB = percent full thickness body surface area burned LOS = length of hospital stay in days APACHE II = acute physiology and chronic health evaluation Il score FLAME = Fatality obtained by Longevity, APACHE II, and Measured Extent of burn

Page 41: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 10 Differences h Non-SPnivon by Gender F

ALL NON-SURVIVORS MEN WOMEN p Value

Patients [n (%)] Ac (Mean years SD) Etbiogy of themil injury

Flame [n (%)] Scald [n (%)] ûther [n (%)]

TBSA (Mean % + SD) FTB (Mean % * SD) lahalition injury [n (%)] N d for mecbuiicil ventilation [n (O!)] Mechinical ventilation (Mean days * SD) Number of opentionr (Mean * SD) LOS (Mean days * SD) Day 1 APACHE II (Mean SD) FLAME (Mean k SD)

p Value * = cornparison between men and women (chi-square test or t-test) OtLcr = electrical, contact, tar, chernical, lightning, and radiation TBSA = percent total body surface area bumed FTB = percent full thickness body surface area bumed LOS = length of hospital stay in days APACHE II = acute physiology and chronic health evaluation II score FLAME = Fatality obtained by Longevity, APACHE II, and Measured Extent of burn

Page 42: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 43: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Table 12 Multiple Logistic Remion Models (n = 527)

Schwartz aLGF ROC -2 LogL pvdue* Criterion p Vdue area f SE

- - - -- --

FLAME 183 0.49 O.% I 0.01 170.58

FLAME + Centres 193 0.57 0.97 f 0.01 155.35 0.004

FLAME + Centres + Sex 194 O. 54 0.97 f 0.01 150.12 0.00 1

FLAME + Centres + Sex + Etiology 237 0.47 0.97 f 0.01 142.96 0.0 1 0

FLAME + Centres+ Sex + Etiology + Inhalation Injury 243 0.27 0.97 i 0.01 142.58 0.0 14

FLAME + Centres+ Sex + Etiology + Inhalation Injury + Ventilation Days 247 0.29 0.97 * 0.01 140.78 0.0 1 3

FLAME + Centrest Sex + Etiology + Inhalation lnjury + Ventilation Days + Delayed Days 247 0.25 0.98 * 0.01 134.1 1 0.003

HLGF = Hosmer-Lemeshow goodness-of-fita ROC = Receiver ûperating Characteristic Cwve SE = Standard Error -2 Log L = -2 Log likelihood ratio pValue* = Chi-square comparing -2 Log Likelihood ratio between FLAME to each of the other models FLAME = Fatatity obtained by Longevity, APACHE II, and Measured Extent of burn

Page 44: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

F Table 13 Two by two diagooltic table for tbe FLAME Score (muiti-centn validation, 1997-1998) (n = 527) nitb

dechion criteria of 0.5,0.7, and 0.9 1 P

Predicteà r isk 0.5 Preüicted rhk 0.7 Predicted rbk 0.9 1.

r'

Observed Observed Observeci 7-

Preàîcted Dead Alive Preàicted Dead Aiive Predicted Dead Alive

Sensitivity Specificity Correct Classification False positive False negative Positive predictive value Negative predicted value Positive Likelihood ratio

- -

FLAME = Fatality obtained by Longevity, APACHE II, and Measured Extent of burn

Page 45: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

A A A

800- - - w u

vv+ ' - V P ? d m - m - d

Page 46: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 47: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

~ o q c q ' Q c y c Y r o d o o o o o o o d

Page 48: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 49: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent
Page 50: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Figure 4 FLAME Score ROC Cuwe

Multi-Centre Validation Population (n = 527)

Area under ROC = 0.96

False Positive

Page 51: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Figure 5 ROC cuwes

False Positive

- 9 - 9

Area under ROC = 0.98

\ Multi-Centre Validation -- \ A T , - -

- ,

-.

-

I I I 1 I I I I I I I I I I I I I 1 I

I

O al û2 a3 a4 Q5 a6 a7 a9 1

Page 52: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

Figure 6 Adult Burn Admissions - 1999 Cross Canada Survey

Page 53: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

-- CHAPTER 5 D Ï ~ Ü ~ s ~ o N -

m e pupose of this chapter is to:

1. sumrnorlte results of the FLAME score multi-centre validation

2. compare l e FLAME score to previous scores

5.1 FLAME score multi-centre validation

With this prospective multi-centre study we have tested and detemined the predictive

ability of the FLAME score among adult burn patients h m five different Canadian bum

centres, using hierarchical multiple logistic regression mode!s, two by two diagnostic

table analysis, area under ROC curve, and the Brier score. Our main finding was that the

FLAME score provides accurate predictions of patient risk of dying during

hospitalization.

5.2 Cornparisons of the FLAME score to previous scores

5.2.2 Similari fies

5.2.2. I Age and percent TBSA

Our results confirm the findings of previous reports, which established the

importance of age, and p e n t TBSA, as important risk factors for mortality in burn

patients?"*'O

Page 54: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

S. 2.1.2- Gender &.A - -

Our results identifi fernale gender as an independent preûictor of mortality and

agree with those of other a ~ t h o r s . * * ~ ~ ~ ' ~ * ' ~ ~ Clark and ~ m m m " have speculated that this

difference may result h m the reduced muscle mass of females in comparison to men

with a subsequent decrease in fuel reserve (e.g., protein and water). One factor that could

explain the increased mortality in our fernale population compared with that of our male

population (18% vs.l3%, p= 0.082, Table 9) may be due to the fact that women were

older than men (49 years vs. 41 years, p c 0.001, Table 9), since older people have

atrophic skin that leads to a larger and deeper bum for any given heat load. I 8

It should be noted that our female population had a higher incidence of scald

bums than that seen in males (23% vs. 13%, p < 0.00 1, Table 9). This type of bum is

o f h indeterminate and requires daily inspection to detemine its full extent. Hence, scald

bums can result in a surgical treatment delay compared to men who only received scald

bums 13% of the time.

5.2.2 D~retences

52.2. I Full thickness burn

Our results did not support 0 t h studies that daim that percent of full thickness

bum (FTB) predicts 0, definition of FTB was the presence of

FTB (any percentage) on admission, diagnosed by the admitting physician (resident, burn

fellow or attending burn surgeon). In several studies the authors have stated that it is

difficult to assess FTB early in the patient's adrni~sion?~*'~~'~"' A major source of

Page 55: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

difficulty in accurately assessing the p-nce of FTB is that over the course of treatment - - u s &A--- - - - -- - - - -

partial thickness burns may evolve, and convert into FTB.'"'~ Moreover, the judgment of

percent FTB is sometimes subje~tive.'~

5.2.2.2 Smoke Inhalation Injury

A number of previous studies have found smoke inhalation injury to be an important

pred ictor of mortdity~ 8202da~3-36.42.n-7s we found no such relationship, yet the number of

patients diagnosed with smoke inhalation injury was relatively low (23%). Our definition

of smoke inhalation injury was based on clinical findings (e.g., bumed in a closed space,

facial burns, sooth in mouth) confirmed within the first 24 hours after burn injury by

bronchoscopy (e.g., sooth in branchial tree, mucosal erythema, edema or ulceration, and

submucosal hemorrhage)." and higher fluid rquirements than those calculated with the

modifieci Parkland formula.

Similar to our results, Vico and Papillon1' found that smoke inhalation injury had no

additional discriminatory power, which nlay imply that al1 the information relating smoke

inhalation injury to bum moitality was incorporated into the percent total body surface

area bumed variable. Saffie and associates4* using Iogistic regression calculations showed

bum patients with smoke inhalation injury had 19 times greater risk of dying, compared

with patients without smoke inhalation injury. However, patients with smoke inhalation

injury suffered larger burns than patients without smoke inhalation injury (32% TBSA vs.

12% TBSA). When burn size and a number of other clinical and demographic variables

were held constant in a multiple logistic regression anaiysis, the increaseà risk of dying

Page 56: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

for-patients with smoke inhalation in& dropped to seven times compared to patients

without smoke inhalation injury.

Smoke inhalation injury was not a significant predictor of mortality in this study.

Maybe because the physiologic derangement produced by this lung injury has been

already measured by the acute physiologic component (e.g., respiratory rate, FiO,, arterial

pH) of the APACHE II score, which is part of the FLAME score.

Smith and as~ociates*~ found smoke inhalation injury to be less significant than

percent TBSA and age. They attributed this to the finding that patients with smoke

inhalation injury were older and more extensively bmed than other patients, and thus at

higher risk of death.

It should be noted that both Saffle and associates4* and Smith and associa te^:^ did

not always diagnose smoke inhalation injury with bronchoscopy, and found an incidence

of smoke inhalation injury lower than expected. Perhaps the use of a "gold standard"

evaluation and an improved definition of smoke inhalation injury in future prospective

studies may f i d e r define its role as an outwme predictor.

5.2.2.3 Multiple Orgon Fuilure

Burn patients, similarly to other critically il1 patients, develop multi-system organ

failUTe.33-~s.n.74 However, bum patients differ h m other critically il1 patients in that they

often are admitted with few initial organ failures. '4243u6*77*78 Thus, one might think that

prognostic indicators assessed early, such as the APACHE II score which is assessed

after the first 24 hours pst-admission, may not be useful since the greatest physiologic

Page 57: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

dermgetnent -develops after several-days pst-bum. In validating the APACHE II score, -- -,A----- - - - - -

Knaus and as~ociates~~ scored patients on ICU and fond the first 24 hours to be the most

useful. hdeed, they found that little change occurred in the score (k 5 points) for al1

patients with a death outcome. They felt it was more appropriate to pertbnn the test over

the initial 24 hours as this score would be the most independent from treatment. We

observed a similar result in our population.

5.2.2.4 Other multivariable scores

Although the APACHE II score has been used previously to test the ability of

other predictive formulas in bum patients,'SJ96' it was not part of those predictive

formulas. The FLAME score is the first bum outcome predictive formula that

incorporates the APACHE II score into its formula. The advantages of adding the

APACHE II score to age and percent TBSA in the FLAME score are that the APACHE 11

score measures the severity of illness (acute physiology score), assess patient's pre-burn

organ dysfùnction (chronic health score), and reinforces the importance of patient's age

(age score) in bum mortality (Table 1).

Page 58: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

CHAPTER 6

CONCLUSIONS

The purpose oJthk chupter is to:

1. sunimarire the main conclusions of the studv

2. explore the implications of the study results

6.1 Main conclusion

This prospective multi-centre study helps to validate the accuracy of the outcome

predictive ability of the FLAME score in adult burn patients Grom different Canadian

bum centres.

6.2 Implications

The FLAME score is an early outcome predictive score that may be used after the

first day of bum unit admission to detennine patient requirernents of nursing care,

diagnostic modalities, medical treatment, surgical interventions, and rehabilitation

therapy. Therefore, the FLAME score may contribute to better allocation of human and

technoIogical resources, triage and management decisions, evaluation of new diagnostic

and treatment modalities, and corn parison of bum populations. Furthemore, the FLAME

score might help in communication between health professionals regarding therapeutic

decisions based on patient mortality risks.

Page 59: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

CHAPTER 7

LIMITATIONS

The purpose of this chapter is to:

I . discuss the limitations of the stuùy

7.1 Few Participating Bum Centres

After sending a f o d written invitation to participate in the study to dl medical

directors of Canadian burn centres: we only obtained data fiom five of them (1 8.5%).

Many of the non-participating burn center directors claimed that they did not have a

person that could collect and send the required data to our burn center. Other medical

directon explained that they admitted few patients per year who were eligible for this

study. Only one medical director declined to participate for personal reasons.

The number of adult patients admitted to the five participating Canadian bum centers

in this study (n = 527) are similar to the number (n = 523) of adult patients admitted to

the Canadian burn centers which completed the annual survey of the Canadian Special

Interest Group of the American B m Association in 1999 (Figure 6). Therefore, the

sarnple size of this study is reasonable, but may not fully reflect the experience at smaller

or les-academically oriented bum centers.

7.2 Single Country Study

The study population corresponds to adult bum patients adrnitted only to Canadian

bum centres. Canada bas a unique health care system with universal patient access, a

Page 60: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

- - - - - -single govemment payor system, and a predetermined physician fee reimbursement

method, which are not similar in other countries. Despite this unique health care system,

the results of the FLAME score in this study could be used for outcome prediction in

bum patients fiom other populations, by comparing patients with similar characteristics

(age, percent TBSA, and APACHE II score).

7.3 Quality of Life

The only outcome that we considered for the FLAME score was survival to discharge

h m hospital. We did not evaluate the quality of life following burn injury in this

Canadian burn population, nor did we examine long-tenn mortality. It is well known that

hypertrophie scars, contractures, and physical limitations may affect the quality of life of

bum survivors. The circumstances of the bum injury incident, unplanned hospitalization

with painful dressing changes, multiple surgeries, several complications, and the

rehabilitation process, may produce an overwhelming psychological trauma.

After hospital discharge, burn sumivon may focus on permanent scars in visible

areas (e.g., face, hands), combined with their inability to pedorm independently activities

of daily living (e.g., dressing, eating, taking a bath), or to perform activities of their

previous job, may produce psychological negative effects that will affect their quality of

life. A better understanding of these long-terni sequelae is a priority for future research.

Page 61: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

7.4 Ethical Considerations -

In this study, we did not evaluate the ethical implications of the utilization of the

FLAME score when considered by the attending physician making critical clinical

decisions. One possible clinical situation is when the burn center beds are al1 occupied

and it is necessary to transfer a burn patient to another hospital ward or home. Another

possible clinicai situation is when it is considered futile to continue aggressive treatment

to a patient (e.g., massive bum injury, multiple organ failure), and it is necessary to make

the decision to withdraw heroic treatment and to provide only cornfort measures. The

accmcy and acceptability of the FLAME score for these and other applications remains

to be explored.

7.5 Adult population

The FLAME score was developed and validated in adult bum patients. Therefore, we

do not know the accuracy of the outcome predictive ability of the FLAME score in

children with bum injuries.

7.6 Self-selection of participating burn centers

It is possible that including only those burn centers which agreed to participate in the

study, we w d d have a selection bias. We do not know the accuracy of the FLAME score

in bura patients from those bum centers which did not participated in the study.

Page 62: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

The purpose of this chapter is to:

1. provide some directionr for fiiure work

8.1 Tests in other populations

It may be appropriate to conduct a rnulti-centre validation of the FLAME score in

adult burn patients fiom other countries with different health care systems than the

Canadian system. It may also be appropnate to test the ability of the FLAME score to

predict outcome in children with bum injuries. In addition, fbrther exploration of adding

o k patient variables (e.g., smoke inhalation) to the FLAME score merits consideration.

8.2 Explore ethical implications

It would be relevant to study the ethical implications of the utilization of the FLAME

score as a CO-adjuvant clinical elernent when the attending physician must make cntical

decisions. In addition, some qualitative assessments by patients, families, and other

riealth care professionais ment consideration.

8.3 Explore ecunomic issues

Prospective studies are needed to evduate the economic implications of using the

FLAME score for the allocation of human and technical resources of bum centers

according to the patient's mortality risk estimated by the FLAME score (page 9).

Page 63: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

A - REFERENCES

1. Bezuhly M, Gomez M, Fish JS. Back to Basics VIT: Minor Bums Injuries. University

of Toronto Medical Journal 2001; 78:232-237.

2. Snelling CF. Burn Units' share of Canada's total burn care. J Burn Care Rehabil 1995;

16:s 19-524.

3. Bull JP, Squire JR. A study of mortality in a bums unit: standards for the evaluation

of alternative methods of treatrnent. Ann Surg 1949; 130: 160- 173.

4. Rittenbury MS, Schmidt FH, Maddox RW, Beazley W, Ham WT, Haynes BW.

Factors significantly affecting mortality in the burned patient. J Trauma 1965; 5587-

600.

5. Rittenbury MS, Maddox RW, Schmidt FH, Ham WT, Haynes BW. Probit analysis of

bum mortality in 183 1 patients: cornparison with other large series. Ann Surg 1966;

164: 123- 138.

6. Stem M, Waisbren BA. Cornpanson of methods of predicting bum mortality. Burns

1978; 6:119-123.

7. McCoy SA, Micks DW, Lynch JB. Discriminant function probability mode1 for

predicting survival in bumed patients. JAMA 1968; 203: 128- 1 30.

8. Moores B, Rahman MM, Browning FSC, Settle JAD. Discriminant function analysis

of 570 consecutive bums patients admitted to the Yorkshire Regional Bums Centre

between 1966 and 1973. Burns 1974; 1 : 135- 14 1.

9. Feller 1, Flora ID, Bawol R. Baseline results of therapy for burned patients. SAMA

1976; 236: 1943- 1947.

Page 64: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

. *LO.-Clark GM, Volenec FJ, Mani MM, Robinson .DW, Humprey W. Predicting the

suMval of bumed patients using discriminant function analysis. Bums 1978; 4:8 1-85.

Il. Tobiasen J, Hiebert JM, Edlich RF. The Abbreviated Burn Severity Index. AM

Emerg Med 1 982; 1 1 :260-262.

12. Tobiasen J, Hiebert M, Edlich RF. A practical b u m severity index. S Bum Care

Rehabil 1982; 3:229-232.

13. Tobiasen J, Hiebert JM. Prediction of burn mortality. Surg Gyn & Obst 1982;

154371 1-714.

14. Roi LD, Flora JD, Davis TM, Wolfe RA. Two new burn severity indices. J Trauma

1983; 23: 1023- 1029.

15. Raff T, Gennann G, Barthold U. Factors influencing the early prediction of outcome

h m burns. Act Chir Plast 1996; 38: 122-1 27.

16. Gemiann G, Barthold U, Lefenng R, Raff T, Hartmann B. The impact of risk factors

and pre-existing conditions on the mortality of burn patients and the precision of

predictive admission-scoring systerns. Burns 1997; 23 : 195-203.

17. Vico P, Papillon J. Factors involved in bum mortality: A multivariate statistical

approach based on discriminant analysis. Bums 1992; 18:2 12-2 15.

1 8. Clark WR, Fromm BS. Burn mortality. Experience at a regional burn unit, litetatw

review. Act Chir Scand 1987; Supp. 537: 1 - 126. 19. Zawacki BE, Azen SP, Imbus SH, Chang YTC. Multifactonal probit analysis of

mortality in burned patients. Ann Surg 1979; 189: 1-5.

Page 65: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

9.26- G, Schqmper M, Kyral E, - Meissl - - O. - - Cornparison - of propostic indices for bums

and assesment of th& accunicy. Burns 1992; 1 8: 1 O9- 1 12.

21. Berry CC, Wachtel TL, Frank HA. An analysis of factors which predict mortality in

hospitalized bum patients. Bums 1982; 9:38-45.

22. Bowser BH, Caldwell FT, Baker JA, Walls RC. Statistical methods to predict

morbidity and mortality : sel f-assessrnent techniques for burn units. Bums 1 982;

9:3 18-326.

23. Ryan CM, Schoenfeld DA, Thorpe WP, Sheridan RL, Caseem EH, Tompkins RG.

Objective esthates of the probability of death from burn injuries. N Engl J Med

1998; 338:362-366.

24. Clark CJ, Reid WH, Gilmour WH, Campbell D. Mortality probability in victims of

fin trauma: revised equation to include inhalation injury. Br Med J 1986; 292: 1303-

1305.

25. Thompson PB, Hemdon DN, Traber DL, Abston S. Effect on mortality of inhalation

injury. .i Trauma 1986; 26: 163- 165.

26. Shirani KZ, Pruitt BA, Mason AD. The influence of inhalation injury and pneumonia

on burn mortality. AM S w g 1987; 20582-87.

27. Smith DL, Cairns BA, Ramadan F, et al. E f f i t of inhalation injury, bum size and age

on mortality: A study of 1477 consecutive bum patients. J Trauma 1994; 37:655-659.

28. Herndon DN, Langner F, Thompson P, Linares HA, Stein M, Traber DL. Pulmonary

injury in burned patients. Surg Clin North Am 1987; 67:3 1-46,

Page 66: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

- 29. Cymri PW, Lutemian A, Braun DW. Shires GT. Bum injury: analysis of survival

and hospitalization time for 937 patients. Ann Surg 1980; 192:472-477.

30. Petenon VM, Murphy JR, Haddix T, Ford P, Anderson SJ, Bartle W. Identification

of novel prognostic indicators in bumed patients. J Trauma 1988; 28:632637.

31. Scott-Conner CEH, Meydrech E. Quantitation of rate of wound closure and the

prediction of death following major burns. Burns 1988; 14:373-378.

32. Cas1 MT, Coen D, Simic D. Senim amyloid A protein in the prediction of postbum

complications and fatal outcome in patients with severe burns. Eur J Clin Biochern

1996; 34:3 1-35.

33. Aikawa N, Shinozawa Y, Ishibiki K, Abe O. Clinical analysis of multiple organ

failure in burned patients. Burns 1987; 13: 103- 109.

34. Marshall WG, Dimick AR. The natural history of major bums with multiple

subsystern failure. J Trauma 1983; 23: 1 02- L 05.

35. Saffle JR, Sullivan J, Tuohig GM, Larson CM. Multiple organ failure in patients with

thermal injury. Crit Care Med 1993; 2 1 : 1673-1 683.

36. Tredget EE, Shankowsky HA, Taenun TV, Moysa GL, Alton JDM. The role of

inhalation injury in burn trauma - A Canadian experience. AM Surg 1990; 2 12:720-

727.

37. McNeil BJ, Keeler E, Adelstein SJ. Primer of certain elernents of medical decision

making. N Engl J Meû 1975; 293:2 1 1-2 15.

38. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating

characteristic (ROC) curve. Radiology 1982; 143:29-36.

Page 67: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

-39,Schuster DP. Predicting outforne after ICU admission. The art and science of

assessing nsk. Chest 1992; 102: 186 1 - 1 870.

40. O'ICeefe GE, Hunt JL, Purdue GF. An evaluation of risk factors for mortdit'y afler

bum trauma and the identification of gender-dependent differences in outcomes. J

Am Col1 Surg 200 1 ; 192: 1 53- 160.

41. Baxter CR. Fluid volume and electrolyte changes of the e d y postbum penod. Clin

Plast Surg 1974; 1 :693-709.

42. Saffle JR, Davis B, Williams P, et al. Recent outcomes in the treatment of burn injury

in the United States: A report h m the American Bum Association patient registry. J

Bwn Care Rehabil 1995; 16:2 19-232.

43. Mann R, Heimbach D. Prognosis and treatment of bums. West J Med 1996; 1652 15-

220,

44. Nguyen TT, Gilpin DA, Meyer NA, Herndon DN. Current treatment of severely

bumed patients. Ann Surg 1 996; 223: 14-25.

45. Hopper RA, Knighton J, Fish J, Peters W. Use of skin substitutes in adult Canadian

bum centres. Can J Plast Surg 1997; 5: 1 12- 1 1 7.

46. Knaus WA, Draper EA, Wagner DP, Zimmeman JE. APACHE II: A severity of

disease classification system. Crit Care Med 1985; l3:8 1 8-829.

47. Knaus WA, Draper EA, Wagner DP, et al. An evaluation of outcome fiom intensive

care in major medical centers. Ann Int Med 1986; 1 O4:4lO4 1 8.

Page 68: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

48. Wong DT, Croh SL, Gomg M, McGuire-GP, Byrjck RJ. Evaluation of predictive --=A- -

ability of APACHE II system and hospital outcome in Canadian intensive care unit

patients. Crit Care Med 1995; 23: 1 177-1 183.

49. Wong DT, Knaus WA. Predicting outcome in critical care: The current status of the

APACHE prognostic system. Can J Anaesth 1 99 1 ; 38:374-383.

50. Seneff M, Knaus WA. Predicting patient outcome fiom intensive care: A guide to

APACHE, MPM, SAPS, PRISM, and other prognostic systems. Intensive Care Med

1990; 5:33-52.

51. Wong DT, B m w PM, Gomez M, McGuire OP. A comparison of the Acute

Physiology and Chronic Health Evaluation (APACHE) 11 score and the Trauma

Injury Severity Score (TRISS) for outcome assessrnent in intensive care unit trauma

patients. Crit Care Med 1 996; 24: 1642- 1647.

52. Kmus WA, Le Gall JR, Wagner DP, et al. A comparison of intensive care in the USA

and France. Lancet 1 982; ii:642-646.

53. Oh TK, Hutchinson R, Short S, et al. Verification of the Acute Physiology and

Chronic Health Evaluation scoring system in a Hong Kong intensive care unit. Crit

Care Med 1993; 2 1 :698-705.

54. Zrowan KM, Kerr JH, Major E, et al. Intensive Care Society's APACHE II study in

Bzitain and Ireland. II: Outcome cornparisons of intensive care units after adjustment

for case mix by the Arnerican APACHE II method. Br Med J 1993; 307:976-98 1.

Page 69: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

55.2-immepan JE, Knaus WA, JudenJA, et al. Patient selection for intensive care. A

comparison of New Zealand and United States hospitals. Crit Care Med 1988;

16:3 18-326.

56. Sirio CA, Tajirni K, Knaus WA, et al. An initial comparison of intensive care in

Japan and the United States. Crit Care Med 1992; 20: 1207- 12 15.

57. Chen FG, Koh KF, Goh MH. Validation of APACHE II score in a surgical intensive

care unit. Singapore Med J 1993; 34:322324.

58. Giangiuliani G, Mancini A, Gui D. Validation of a severity of illness score

(APACHEII) in a surgical intensive care unit. Intensive Care Med 1989; 155 19-522.

59. Cannon JO, Friedberg JS, Gelfmd JA, Tompkins RG, Burke JF, Dinarello CA.

Circulating interleukin-lp and tumor necrosis factors concentrations after burn

injury in humans. Cnt Care Med 1992; 20: 14 14- 14 19.

60. Brown PE, McClave SA, Hoy NW, Short AF, Sexton LK, Meyer KL. The acute

physiology and chronic health evaluation II classification system is a valid marker for

physiologie stress in the criticall y il1 patient. Crit Care Med 1 993 ; 2 1 :363-367.

61. Viscardi PJ, Polk HCJ. Outcome of amputations in patients with major bms. Burns

1995; 2 1 526-529.

62. Gomez M, Stewart G, Petm W, et al. Improved prediction of mortality in burn

patients. Clin hvest Med 1996; 19:s 13.

63. Lemeshow S, Hosmer DW. A review of goodness of fit statistics for use in the

development of logistic regression models. Am J Epiderniol 1982; 1 1 S:92- 106.

Page 70: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

A .W-V&ecchi MG, Mode- therelative riskaf esophageal cancer in a case-control

study. J Clin Epidemiol 1 992; 45347-355.

65. Redelmeier DA, Bloch DA, Hickam DH. Assessing predictive accuracy: How to

compare Brier scores. J Clin Epidemiol 199 1 ; 44: 1 14 1 - 1 146.

66. Cody RP, Smith JK. Multiple Logistic Regression Analysis. In: River US, ed.

Applied statistics and the SAS programming language. New Jersey: Prentice Hall

Inc., 1 997:235-246.

67. Sackett DL, Haynes RB, Guyatt GH, Tugwell P. nie interpretation of diagnostic data.

Clinical Epidemiology - A basic science for clinical medicine. Toronto: Little Brown

and Company, 199 1.

68. GLIM System. Oxford: The numerical algorithm Group, 1985.

69. SAS/STAT User's Guide. Cary, NC, USA: SAS Institute Inc, 1990.

70. Bull JP, Fisher Al. A study of mortality in a bums unit: a revised estimate. AM Surg

1954; 139:269-274.

7 1. Bull JP. Revised analysis of mortality due to bums. The Lancet 197 1; 2: 1 133- 1 134.

72. Saffle JR, Larson CM, Sullivan J, Shelby J. The continuing challenge of burn care in

the elderly. Surgery 1 990; 1 O8:534-543.

73. Rue LW, Cioffi WG, Mason AD, McManus WF, Pruitt BA. Improved survival of

bumed patients with inhalation injury. Arch Surg 1993; 128: 772-778.

74. Tweed A, Ross JF. A review of the mortality in the b m units at the Victoria General

Hospital and the Izaak Walton Killam Hospital, January 1967 to April 1977. Ann

Plat S ~ r g 1979; 2349 1-498.

Page 71: Predicting Death in Burn Patients DEATH IN BURN PA'i'RZNTS - Master ... Multiple logistic regression anaiysis O 1990 ... Burns are devastating injuries that can produce permanent

2SJmitt B k Cioffi-WG, Shimani TF Ikeuchi H, Mason AD. Evaluation and

management of patients with inhalation injury. J Trauma 1990; 30:S63-S68.

76. Hunt IL, Agee RN, Pruitt BA. Fiberoptic bronchoscopy in acute inhalation injury. J

Trauma 1 97 5; 1 5:64 1 -649.

77. Knaus WA, Zimmerman JE, Wagner DP, Draper EA, Lawrence DE. APACHE - acute physiology and chronic health evaluation: a physiologically based classification

system. Cnt Care Med 198 1 ; 959 1-597.

78. Marshall IC, Cook DJ, Christou NV, Bernard GR, Sprung CL, Sibbald WJ. Multiple

Organ Dyshction Score: A reliable descriptor of a complex clinical outcorne. Crit

Care Med 1995; 23: 1638-1652.