Annals of the MBC - vol. 4 - n' I - March 1991

DEATH PROBABILITY EVALUATION IN CRITICALLY BURNED PATIENTS WITH A MULTIVARIANT ADJUSTMENT

Herruzo Cabrera R., Gil Miguel A., Garcia Torres J., Rey Calero J., Mayer Fournaraki R.F.

Preventive Medicine and Public Health Department, Medicine Faculty, Autonomous University of Madrid, Madrid, Spain


SUMMARY. 193 critically burned patients (over 25% damaged body surface or affected by electric mechanisms) under treatment in the Burned Patients Unit of La Paz Hospital from 1985 to 1987 were examined and their death probability was calculated through a logistic regression model, which included their clinical characteristics (age, burned body surface, inhalation syndrome, delayed eschar debridement, previous pathology, etc.). All variables were included in the Logress program, which establishes a model with the principal predictive value of death. The model was evaluated through a Characteristics Operating Curve, where the optimum cut point was located at a 55% theoretical probability of death, with a sensitivity of 82%, a specificity of 82%, and Positive Predictive Value of 60.6%. The Negative Predictive Value was 93%, and the Likelihood Ratio 4.5.

Introduction
The bum patient is the prototype of an immunodepressed patient, especially during the first days after the accident, presenting a high risk of infection and death (1, 2, 3). These risks are described in connection with two principal factors, which are the extent and the depth of the bums (1, 2, 4, 5, 6). However, when only patients with a greatly extended burned surface are examined ( ~! 2 5% body surface) or those affected by a special burning mechanism (electricity) including the so-called critically burned patients, there are other factors which may also affect the outcome. These. factors include previous pathology, inhalation syndrome, and the handling occurring while in the Bum Unit, as we have mentioned in a previous descriptive study published in this journal (7). In this study we intend to establish a predictive model concerning the probability of death in critical bum patients, using the significant risk and protective factors (RF, PF) and to investigate their relative effect on the death risk of the patient.

Materials and methods
A study was made of 193 critically burned patients in the Burned Patients Unit of La Paz Hospital during the years 1985-1987. For this purpose a special epidemiological card was developed including the essential data. The following information was included: age, sex, extent of bum, type of bum, the interval between the bum and treatment, time of eschar debridement, type of debridement, inhalation syndrome, previous pathology, complications, death and its cause.
The statistical analysis was carried out in two stages: a univariable analysis which selected those significant variables that were' then included in a logistic model; and the selection of variables through a backward stepwise procedure (8). All the variables were dichotomized, taking 2 values of 0 and I except in the case of the age and body surface which were qualified in four values. By this manner the 14 variables used in the first stage were augmented to 35 in a later classification (Tab. 1).

Results
Of the 35 variables, only 6 were statistically significant, 5 of them as risk factors and one as a protection factor. The risk factors were age, electric bum, extent of bum, inhalation syndrome and previous pathology. The protection factor was the time of the eschar debridement (Tab. 2).
We can calculate the probability of death through the equation or, following an easier way, through a conversion of products "coefficient x categories" (from each variable) into "weights" (9), as we can see in Tab. 3, obtaining a sum of these weights + the constant of the equation, and finally, transforming this value (total sum) into a probability through the conversion table (Tab. 4).
Example: a patient (F.M.P.) was admitted to the Burned Patients Unit, presenting with the following clinical characteristics:

  1. Age 43 y. (value = 0.42, see Tab. 3)
  2. Body surface burned 75% (value = 5.12, see Tab, 3) ,
  3. Previous pathology YES (value = 1.72, see Tab. 3)
  4. Inhalation Syndrome YES (value = 2.5, see Tab. 3)
  5. Burning mechanism Others (value = 0, see Tab. 3) 0 Debridement < 8 d. (value = 0, see Tab. 3) g) Constant (value = - 5.37, see Tab. 3)

Tab. 1 Relation of the studied variables

  1. Age (0--16-35, 1=:515, 2=36-55, 3=56-70, 4= >70 Years)
  2. Sex (0--Male, I=Female)
  3. Body burned surface (0_-;SI0%, 1=1-30%, 2=31-50%, 3=51-70%, 4=>70%)
  4. Burning mechanism (0--Scald, I=Electric)
  5. I=Flame)
  6. I=Explosives)
  7. 1---Chemicals)
  8. I=Low fire)
  9. First day of eschar debridement (0-- ~-- 3, 1---4-7)
  10. (0-5 3, 1=8-15)
  11. (0-z3, 1=>15)
  12. Type of eschar debridement (0--Tangential, I=Fascial)
  13. (0--Tangential, I=Mixed)
  14. Day of eschar debridement (0--z~7, 1=8-15)
  15. (0--;57, 1=16-25)
  16. (0-- --s 7, 1 = > 2 5)
  17. Inhalation syndrome (0--No, I=Yes)
  18. Day of mechanic ventilation (0_-:i7, 1=8-15)
  19. (0-7, 1=16-25)
  20. (0--~ 7, 1 = > 2 5)
  21. Previous pathology (0--No, I=Yes)
  22. Complications of the bum (0--None, I=Sepsis)
  23. (0--None, I=Local infection)
  24. (0--None, 1=lkeSpiratory infection)
  25. (0--None, 1---Cardiopathy)
  26. (0--None, I=Respiratory distress)
  27. (0--None, 1---Others)
  28. Day of complete recovering (0, 1=1 1-20)
  29. (0-~ 10, 1=21-30)
  30. (0-510, 1= >30)
  31. Death (0--No, I=Yes)
  32. Cause of death (0--Others, I=Sepsis)
  33. (0--Others, 1=Respiratory distress)
  34. (0--Others, 1--Cardiopathy)
  35. (0--Others, I=Metabolic distress)

This calculation can be easily made with a manual calculator or better with an electronic spreadsheet in a PC. We previously elaborated a number of calculations which represent the weights of the categories of the model variables (Tab. 3). Then we added them together in different examples and calculated the expected probability of death for a range of sums of the weights (Tab. 4).
In our example, the sum of weights for the patient is 3.9 and according to Tab. 4 the expected probability of death is 98%. We suggest that physicians in Bum Units carry Tab. 4 with them in order to facilitate the calculation of death probability in clinical practice.
Finally, concerning the probability results for each patient, we can estimate the internal validity of this equation, as a simple test in critical bum patients. For this we elaborate the ROC for these patients (10 ' 11, 12) obtaining the optimum point where the predictive values and the LR (likelihood radio) of the equation are at their maximum.
The internal validation of our equation indicates an optimum point of the equation in the probability of death of 55% having a sensitivity of 82%, a specificity of 82% with a positive predictive value of 60.6%, and a negative predictive value of 92.9%. The percentage of concordance (in well-classified subjects) was 81.9%, with an LR of 4.5 (p < 0.05).

Discussion

Fig. 1: Curve ROC

 

Independent Variable:

Coefficient

Std. Error

"Z"

age (0, 1, 2, 3, 4)

0.21

0.18

1,13

Mechanical Burn (0=scald, l=electric)

0.90

0.44

2,05

Eschar debridement 1 (0= -,:z 7 or ~> ~ 2 5, 1 =8 -2 5)

-1.5

0.40

-3.74

Inhalation S. (0=no, l=yes)

2.49

0.51

4.9

P. Pathology (0=no, l=yes)

1.72

0.54

3.17

Surface Body Burn (0, 1, 2, 3, 4)

1.28

0.27

4.78

Constant

-5.37

0.86

-6.20

Independent Variat:

Odds ratio

Lower Limit

Upper Limit

age (0, 1, 2, 3, 4)

1.23

0.85

1.78

Mechanical Burn (0=scald, l=electric)

2.48

IM

5.91

Eschar debridement (0= :~ 7 or ~> 2 5, 1 =8 -2 5)

0.22

0,10

0.48

Inhalation S. (0=no, l=yes)

12.14

4.46

33.01

P. Pathology (0=no, l=yes)

5.58

1.92

16.21

Surface Body Burned (0, 1, 2, 3, 4)

3.63

2.13

6.16

Tab. 2 Regression coefficients of the significant variables and their standard error, Z value and the odds ratio with their confidence limits

Dependent Variable: death (0=no, l=yes)1

Deatliprob = ------------------------------------------------------------- = 98%

l+e - (aO +0.2 XI -~ 0-9 X2 + 2.5 X3 + 1.7 X4 + 1.3 x5 ~ 1.5 X6)

aO = - 5,3

XI = Age

x2 = Bum mechanism

x3 = Inhalation syndrome

x4 = Previous pathology

x5 = Bum-body surface

x6 = Eschar Debridement time

Moreover logistic regression permits us to make the predictive model and direct calculation of the odds ratio for each independent variable (controlling the effect of the others) and is thus more useful from the epidemiological point of view (13, 14, 15, 16, 17) to evaluate the effect of each factor and its consideration as RF or PF. In our case we have 5 well defined RF and I PF.
Of these 5 RF, the best known were age and the extent of burned body surface (1, 2, 4, 5, 6). Other authors, such as Zawacki (16), also include previous pathology and the inhalation syndrome. To these we added a significant RF and the type of bum ending with a PF at the time of scar debridement.
On the other hand, due to the complexity that clinicians may find in the application of our formulas, we have transformed them into a simple sum of distinct factors. This value can be converted into a probability through the conversion table (9, 17).
Finally, as an internal method of validation, we insist that the point of greatest value in the global equation, considering this as the only test performed, is a ~! 55% probability of death, while the LR and the concordance percentage (in well-classified subjects) are maximum. However, this point (mathematically established by the bisector between sensitivity and the complementary specificity)_ may be modified by the clinician's criteria, according to what he is trying to establish. If he wants to establish who will die at the expense of losing specificity and increasing the false positives, the threshold will be less than 55%. On the contrary, if he wants to know who will not die, considering the loss of sensitivity and the increase in the number of false negatives, the threshold will be bigger than that established (55%).

RESUME Les Auteurs ont kudié 193 patients critiquement bi-é1és (bridures touchant 25% de la surface corporelle ou causées par 1'é1ectricité), hospitalisés chez I'Unité de Br~1és de I'Hépital La Paz dans la période 1985-87. Ils ont calcuk la probabilité de mort de ces patients avec un modéle de régression logique qui incluait leurs charactéristiques cliniques (dge, surface corporelle brfflée, syndrome d'inhalation, débridement retardé de Pescarre, pathologie précédente, etc.). Toutes les variables ont été inclues dans le programme Logress, qui établit un mo&le avec la valeur prédictive principale de la mort. Les résultats de la recherche sont reportés.


BIBLIOGRAPHY

  1. Polk H.O.: Consensus summary on infection. Journal of Traumatology, 19: 894-895, 1979.
  2. Curreri W.P., Luterman A., Braum D.W., Shirer G.T.: Bum injury: Analysis of survival and hospitalization time for 937 patients. Annals of Surgery, 192: 472-478, 1980.
  3. Gelfand J.A.: Infections in burnt patients: A paradigm for cutaneous infection in the patient at risk. American Journal of Medicine, 76: 158-165, 1984.
  4. Herruzo Cabrera R., Garcia Torres V., Rey Calero J.: Estudio epidemiológico de la infección hospitalaria en una unidad de quemados durante dos años. Revista Espafiola de Microbiología Clínica, 4: 145-150, 1989.
  5. Clark W.R., Fromm B.S.: 1982: Bum mortality. Bulletin Clinical Review Bum Injury, 1: 61-62, 1982.
  6. Bowser B.H., Caldwell F.T.: Probit and discriminant analysis of bum mortality: a protocol for improved survival. Bulletin Clinical Review Bum Injury, 1: 60-61, 1986.
  7. Santos Heredero J., Garcia Torres V., Herruzo Cabrera R., Fernandez Delgado J.: Risk factors in the critical burn patients. Annals of the MBC, 3: 160-165, 1990.
  8. Greenland S.: Modeling and variable selection in epidemiologic analysis. 79: 340-349, 1989.
  9. Sox H.C.: Probability theory in the use of diagnostic tests: an introduction to critical study of the literature. Annals of Internal Medicine, 104: 60-66, 1986.
  10. Goldman L., Whtemanx C., Garfield F.: Impact of cardiology data bank on physicians' prognostic estimates. Archive Internal Medicine, 141: 1631-1636, 1981.
  11. Rey Calero J.: "Metodo Epidemiol6gico y salud de la comunidad". Ed. McGraw-Hill Interamericana, Madrid, 163-169, 1989.
  12. Pauker S.G., Kassirer J.P.: Clinical application of decision analysis: A detailed illustration. Seminary Nuclear Medicine, 302: 1109-1125, 1980.
  13. Rothman K.J.: "Modem Epidemiology". Ed. Little, Brown and Co., New York, 1986.
  14. Press S.J., Wilson S.: Choosing between logistic regression and discriminant analysis. Journal American Statistics Association, 73: 699-705, 1978.
  15. Cupples L.A., Heeren T., Schalzkin A., Colton T.: Multiple testing of hypotheses in comparing two groups. Annals Internal Medicine, 100: 122-129, 1984.
  16. Zawacki B.E., Azen S.P., Imbus S.H., Chang Y.T.: Multifactorial probit analysis of mortality in burned patients. Annals Surgery, 1: 1-5, 1979.
  17. Rembold, C.M., Watson D.: Post-test probability calculation by weights. A simple form of Bayes' Theorem. Annals Internal Medicine, 108: 115-120, 1988.



 

Contact Us
mbcpa@medbc.com