Información de la revista
Vol. 17. Núm. 6.
Páginas 504-511 (Noviembre - Diciembre 2003)
Respuestas rápidas
Compartir
Compartir
Descargar PDF
Más opciones de artículo
Vol. 17. Núm. 6.
Páginas 504-511 (Noviembre - Diciembre 2003)
Open Access
Aplicación de las redes neuronales artificiales para la estratificación de riesgo de mortalidad hospitalaria
Application of artificial neural networks for risk stratification of hospital mortality
Visitas
5773
J. Trujillanoa,
Autor para correspondencia
jtruji@cmb.udl.es

Correspondencia: Hospital Arnau de Vilanova. Unidad de Cuidados Intensivos. Avda. Rovira Roure, 80. 28198 Lleida. España.
, J. Marchb, M. Badiaa, A. Rodrígueza, A. Sorribasb
a Unidad de Cuidados Intensivos. Hospital Universitario Arnau de Vilanova de Lleida. Lleida
b Departamento de Ciencias Médicas Básicas. Universidad de Lleida. Lleida. España
Este artículo ha recibido

Under a Creative Commons license
Información del artículo
Resumen
Objetivo

Comparar la capacidad de predicción de mortalidad hospitalaria de una red neuronal artificial (RNA) con el Acute Physiology and Chronic Health Evaluation II (APACHE II) y la regresión logística (RL), y comparar la asignación de probabilidades entre los distintos modelos.

Método

Se recogen de forma prospectiva las variables necesarias para el cálculo del APACHE II. Disponemos de 1.146 pacientes asignándose aleatoriamente (70 y 30%) al grupo de Desarrollo (800) y al de Validación (346). Con las mismas variables se genera un modelo de RL y de RNA (perceptrón de 3 capas entrenado por algoritmo de backpropagation con remuestreo bootstrap y con 9 nodos en la capa oculta) en el grupo de desarrollo. Se comparan los tres modelos en función de los criterios de discriminación con el área bajo la curva ROC (ABC [IC del 95%]) y de calibración con el test de Hosmer- Lemeshow C (HLC). Las diferencias entre las probabilidades se valoran con el test de Bland-Altman.

Resultados

En el grupo de validación, el APACHE II con ABC de 0,79 (0,75-0,84) y HLC de 11 (p = 0,329); modelo RL, ABC de 0,81 (0,76-0,85) y HLC de 29 (p = 0,0001), y en RNA, ABC de 0,82 (0,77-0,86) y HLC de 10 (p = 0,404). Los pacientes con mayores diferencias en la asignación de probabilidad entre RL y RN (8% del total) son pacientes con problemas neurológicos. Los peores resultados se obtienen en los pacientes traumáticos (ABC inferior a 0,75 en todos los modelos). En los pacientes respiratorios, la RNA alcanza los mejores resultados (ABC = 0,87 [0,78-0,91]).

Conclusiones

Una RNA es capaz de estratificar el riesgo de mortalidad hospitalaria utilizando las variables del sistema APACHE II. La RNA consigue mejores resultados frente a RL, sin alcanzar significación, ya que no trabaja con restricciones lineales ni de independencia de variables, con una diferente asignación de probabilidad individual entre los modelos.

Palabras clave:
Mortalidad hospitalaria
Estratificación de riesgo
Unidad de cuidados intensivos
Redes neuronales artificiales
Bootstrap
Abstract
Objective

To compare the ability of an artificial neural network (ANN) to predict hospital mortality with that of the Acute Physiology and Chronic Health Evaluation II (APACHE II) system and multiple logistic regression (LR). A secondary objective was to compare the allocation of individual probability among the models.

Method

The variables required for calculating the APACHE II were prospectively collected. A total of 1146 patients were divided (randomly 70% and 30%) into the Development (800) and the Validation (346) sets. With the same variables an LR model and an ANN were carried out (a 3-layer perceptron trained by algorithm backpropagation with bootstrap resampling and with 9 nodes in the hidden layer) in the Development set. The models developed were contrasted with the Validation set and their discrimination properties were evaluated using the area under the ROC curve (AUC [95% CI]) and calibration with the Hosmer-Lemeshow C (HLC) test. Differences between the probabilities were evaluated using the Bland-Altman test.

Results

The Validation set showed an APACHE II with an AUC = 0.79 (0.75-0.84) and HLC = 11 (p = 0.329); LR model AUC = 0.81 (0.76-0.85) and HLC = 29 (p = 0.0001) and an ANN AUC = 0.82 (0.77-0.86) and HLC = 10 (p = 0.404). The patients with the most important differences in the allocation of probability between LR and ANN (8% of the total) were neurological. The worst results were found in trauma patients with an AUC of not greater than 0.75 in all the models. In respiratory patients, the ANN achieved the best AUC = 0.87 (0.78- 0.91).

Conclusions

The ANN was able to stratify hospital mortality risk by using the APACHE II system variables. The ANN tended to achieve better results than LR, since, in order to work, it does not require lineal restrictions or independent variables. Allocation of individual probability differed in each model.

Key words:
Mortality
Risk assessment
Intensive Care Unit
Artificial Neural Network
Bootstrap
El Texto completo está disponible en PDF
Bibliografía
[1.]
J. Librero, R. Ordiñana, S. Peiró.
Análisis automatizado de la calidad del conjunto mínimo de datos básicos. Implicaciones para los sistemas de ajuste de riesgos.
Gac Sanit, 12 (1998), pp. 9-21
[2.]
C. Díaz, D. Martínez, I. Salcedo, J. Masa, J. De Irala, R. Fernández-Crehuet.
Influencia de la infección nosocomial sobre la mortalidad en una unidad de cuidados intensivos.
Gac Sanit, 12 (1998), pp. 23-28
[3.]
A. Armoni.
Use of neural networks in medical diagnosis.
MD Computing, 15 (1998), pp. 100-104
[4.]
D.J. Sargent.
Comparison of artificial neural networks with other statistical approaches. Results from medical data sets.
Cancer, 91 (2001), pp. 1636-1642
[5.]
J.E. Dayhoff, J.M. DeLeo.
Artificial neural networks. Opening the black box.
Cancer, 91 (2001), pp. 1615-1635
[6.]
W.G. Baxt.
Application of artificial neural networks to clinical medicine.
Lancet, 346 (1995), pp. 1135-1138
[7.]
D. Axelson, I.J. Bakken, I. Susann, B. Ehrnholm, G. Nilsen, J. Aasly.
Applications of neural network analyses to in vivo 1H magnetic resonance spectroscopy of Parkinson disease patients.
J Magn Reson Imaging, 16 (2002), pp. 13-20
[8.]
M. Ohlsson, H. Ohlin, S.M. Wallerstedt, L. Edembrandt.
Usefulness of serial electrocardiograms for diagnosis of acute myocardial infarction.
Am J Cardiol, 88 (2001), pp. 478-481
[9.]
S. Yamamura, K. Nishizawa, M. Hirano, Y. Momose, A. Kimura.
Prediction of plasma levels of aminoglycoside antibiotic in patients with severe illness by means of an artificial neural network simulator.
J Pharm Sci, 1 (1998), pp. 95-101
[10.]
D.M. Coulter, A. Bate, R.H. Meyboom, M. Lindquist, I.R. Edwards.
Antipsychotic drugs and heart muscle disorder in international pharmacovigilance: data mining study.
BMJ, 322 (2001), pp. 1207-1209
[11.]
R.P. Lippmann, D.M. Shahian.
Coronary artery bypass risk prediction using neural networks.
Ann Thorac Surg, 63 (1997), pp. 1635-1643
[12.]
W.G. Baxt, F.S. Shofer, F.D. Sites, J.E. Hollander.
A neural network aid for the early diagnosis of cardiac ischemia in patients presenting to the emergency department with chest pain.
Ann Emerg Med, 40 (2002), pp. 575-583
[13.]
R.F. Levine.
Conference concluding remarks.
Cancer, 91 (2001), pp. 1696-1697
[14.]
W.A. Knaus, E.A. Draper, D.P. Wagner, J.E. Zimmerman.
APACHE II: A severity of disease classification system.
Crit Care Med, 13 (1985), pp. 818-829
[15.]
J.V. Tu.
Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes.
J Clin Epidemiol, 49 (1996), pp. 1225-1231
[16.]
G.D. Tourassi, C.E. Floyd.
The effect of data sampling on the performance evaluation of artificial neural networks in medical diagnosis.
Medical Decision Making, 17 (1997), pp. 186-192
[17.]
J. Llorca, T. Dierssen.
Comparación de dos métodos para el cálculo de la incertidumbre en los análisis de laboratorio.
Gac Sanit, 14 (2000), pp. 458-463
[18.]
B.S. Cross, R.F. Harrison, R.L. Kennedy.
Introduction to neural networks.
Lancet, 346 (1995), pp. 1075-1079
[19.]
Neural Network FAQ (Sarle WS) [consultado 5/05/2002]. Disponible en: ftp://ftp.sas.com/pub/neural/FAQ.html
[20.]
J.A. Hanley, B.J. McNeil.
The meaning and use of the area under a receiver operating characteristic (ROC) curve.
[21.]
J.A. Hanley, B.J. McNeil.
A method of comparing the areas under receiver operating characteristic curves derived from the same cases.
Radiology, 148 (1983), pp. 839-843
[22.]
S. Lemeshow, D.W. Hosmer.
A review of goodness of fit statistics for use in the development of logistic regression models.
Am J Epidemiol, 115 (1982), pp. 92-106
[23.]
J. Rapoport, D. Teres, S. Lemeshow, S. Gehlbach.
A method for assessing the clinical performance and cost-effectiveness of intensive care units: a multicenter inception cohort study.
Crit Care Med, 22 (1994), pp. 1385-1391
[24.]
J.M. Bland, D.G. Altman.
Statistical methods for assessing agreement between two methods of clinical measurement.
Lancet, 1 (1986), pp. 307-310
[25.]
R. Abizanda, B. Balerdi, J. Lopez, F.X. Valle, R. Jorda, I. Ayestaran, C. Rubert.
Fallos de predicción de resultados mediante APACHE II. Análisis de los errores de predicción de mortalidad en pacientes críticos.
Med Clin (Barc), 102 (1994), pp. 527-531
[26.]
Qnet (Vesta Services Inc.) [consultado 5/05/2002]. Disponible en: http:/www.qnetv2k.com/qnet2000information.htm
[27.]
L.S.S. Wong, J.D. Young.
A comparison of ICU mortality prediction using the APACHE II scoring system and artificial neural network.
Anaesthesia, 54 (1999), pp. 1048-1054
[28.]
G. Clermont, D.C. Angus, S.M. DiRusso, M. Griffin, W.T. Linde-Zwirble.
Predicting hospital mortality for patients in the intensive care unit: A comparison of artificial neural networks with logistic regression models.
Crit Care Med, 29 (2001), pp. 291-296
[29.]
D.J. Sargent.
Comparison of artificial neural networks with other statistical approaches. Results from medical data sets.
Cancer, 91 (2001), pp. 1636-1642
[30.]
B. Martin, A. Sanz.
Redes neuronales y sistemas borrosos, Editorial Ra-Ma, (1997),
[31.]
A. Ciampi, F. Zhang.
A new approach to training back-propagation artificial neural networks: empirical evaluation on ten data sets from clinical studies.
Statist Med, 21 (2002), pp. 1309-1330
[32.]
P.E. Marik, J. Varon.
Severity scoring and outcome assessment. Computerized predictive models and scoring systems.
Crit Care Clin, 15 (1999), pp. 633-646
[33.]
D.Y. Cho, Y.C. Wang.
Comparison of the APACHE III. APACHE II and Glasgow coma scale in acute head injury for prediction of mortality and functional outcome.
Intensive Care Med, 23 (1997), pp. 77-84
[34.]
D.J.J. Muckart, S. Bhagwanjee, E. Gouws.
Validation of an outcome prediction model for critically ill trauma patients without head injury.
J Trauma, 43 (1997), pp. 934-939
[35.]
M. Álvarez, J.M. Nava, M. Rue, S. Quintana.
Mortality prediction in head trauma patients: Performance of Glasgow Coma Score and general severity systems.
Crit Care Med, 26 (1998), pp. 142-148
[36.]
K. Liestol, P.K. Anderesen, U. Andersen.
Survival analysis and neural nets.
Statist Med, 13 (1994), pp. 1189-1200
Copyright © 2003. Sociedad Española de Salud Pública y Administración Sanitaria
Idiomas
Gaceta Sanitaria
Opciones de artículo
Herramientas
es en

¿Es usted profesional sanitario apto para prescribir o dispensar medicamentos?

Are you a health professional able to prescribe or dispense drugs?