M. Badia

Aproximación a la metodología basada en árboles de decisión (CART). Mortalidad hospitalaria del infarto agudo de miocardio

Objective: To provide an overview of decision trees based on CART (Classification and Regression Trees) methodology. As an example, we developed a CART model intended to estimate the probability of intrahospital death from acute myocardial infarction (AMI). Method: We employed the minimum data set (MDS) of Andalusia, Catalonia, Madrid and the Basque Country (20012002), which included 33,203 patients with a diagnosis of AMI. The 33,203 patients were randomly divided (70% and 30%) into the development (DS; n = 23,277) and the validation (VS; n = 9,926) sets. The CART inductive model was based on Breiman’s algorithm, with a sensitivity analysis based on the Gini index and cross-validation. We compared the results with those obtained by using both logistic regression (LR) and artificial neural network (ANN) (multilayer perceptron) models. The developed models were contrasted with the VS and their properties were evaluated with the area under the ROC curve (AUC) (95% confidence interval [CI]). Results: In the DS, the CART showed an AUC = 0.85 (0.860.88), LR 0.87 (0.86-0.88) and ANN 0.85 (0.85-0.86). In the VS, the CART showed an AUC = 0.85 (0.85-0.88), LR 0.86 (0.85-0.88) and ANN 0.84 (0.83-0.86). Conclusions: None of the methods tested outperformed the others in terms of discriminative ability. We found that the CART model was much easier to use and interpret, because the decision rules generated could be applied without the need for mathematical calculations.

Atención del paciente crítico pediátrico en una UCI de adultos: utilidad del índice PIM

OBJECTIVES An analysis is made of the characteristics of patients younger than 14 years treated in an adult ICU (AICU), to determine the procedures and techniques required by such patients, and to evaluate the use of the Pediatric Index of Mortality (PIM) in stratifying severity. DESIGN A retrospective observational study was carried out. SETTING An AICU of a second level hospital. PATIENTS We studied 130 patients aged from 1 month to 14 years (average age 6.1±4 years) treated in the AICU from January 1997 to December 2010. VARIABLES OF INTEREST Clinical-demographic parameters, diagnosis, clinical procedures, PIM score, length of stay, transfer to pediatric ICU (PICU), and mortality. Classification by destination (AICU, PICU) and outcome (alive, dead). PIM and assessment of the diagnostic performance curve (ROC) for mortality. RESULTS The average age of the patients was 6.1±4 years. Most common diagnoses: trauma (26.9%) and sepsis (22.3%). Main procedures: mechanical ventilation (58.5%), central venous line (74.6%) and vasoactive drugs (20%). A total of 64.6% were transferred to PICU, and the overall mortality was 13%. Patients who stayed in the AICU were older (8.2±4 vs 5.5±4 years, p<0.001), had low morbidity, and their stay was short (44.5±38 hours). The PIM score was significantly higher in the patients who died (60±20 AICU, 38±30 PICU) than in those who survived (4±1 AICU, 9±1 PICU) (p<0.001). ROC curve with AUC=0.91 (95%CI: 0.85 to 0.98). CONCLUSIONS The PIM score can stratify severity and identify patients at an increased risk of death. Critical child care in the AICU requires the presence of adequate materials and the continuous learning of procedures adapted to pediatric patients in order to ensure adequate care.

Approach to the methodology of classification and regression trees

OBJECTIVE To provide an overview of decision trees based on CART (Classification and Regression Trees) methodology. As an example, we developed a CART model intended to estimate the probability of intrahospital death from acute myocardial infarction (AMI). METHOD We employed the minimum data set (MDS) of Andalusia, Catalonia, Madrid and the Basque Country (2001-2002), which included 33,203 patients with a diagnosis of AMI. The 33,203 patients were randomly divided (70% and 30%) into the development (DS; n = 23,277) and the validation (VS; n = 9,926) sets. The CART inductive model was based on Breimans algorithm, with a sensitivity analysis based on the Gini index and cross-validation. We compared the results with those obtained by using both logistic regression (LR) and artificial neural network (ANN) (multilayer perceptron) models. The developed models were contrasted with the VS and their properties were evaluated with the area under the ROC curve (AUC) (95% confidence interval [CI]). RESULTS In the DS, the CART showed an AUC = 0.85 (0.86-0.88), LR 0.87 (0.86-0.88) and ANN 0.85 (0.85-0.86). In the VS, the CART showed an AUC = 0.85 (0.85-0.88), LR 0.86 (0.85-0.88) and ANN 0.84 (0.83-0.86). CONCLUSIONS None of the methods tested outperformed the others in terms of discriminative ability. We found that the CART model was much easier to use and interpret, because the decision rules generated could be applied without the need for mathematical calculations.

Skin lesions after intensive care procedures: Results of a prospective study ☆

PURPOSE To define the skin lesions produced by procedures used in the intensive care unit (ICU) and to examine patients 12 months after discharge from the ICU. MATERIAL AND METHODS This was a prospective clinical study in the 14-bed multidisciplinary ICU of a university hospital. Iatrogenic skin lesions (ISL) were examined in 316 patients after ICU discharge. RESULTS A total of 189 patients were interviewed 12 months after ICU discharge. More than 85% of the patients had ISL after being discharged from the ICU. The patients with the highest Acute Physiology and Chronic Health Evaluation II score and longest average stay presented the highest number of ISLs. A total of 93 patients (49%) reported some skin lesions after 12 months. All patients who had undergone surgical tracheostomy reported the presence of a scar, but 4 of 24 patients who had undergone percutaneous tracheostomy reported no tracheostomy scar. Only 22% of all patients reported scars caused by vascular catheter access. About half (54.5%) of the patients reported secondary lesions caused by chest draining, and these were predominantly caused by the large-bore tube drainage. All patients reported the presence of a laparatomy scar. CONCLUSIONS Most patients had identified skin lesions resulting from ICU procedures. Half of all patients were aware of their lesions and reported them at 12 months. Future research is needed to understand whether these lesions cause problems to survivors quality of life and whether the lesions lead to increased health care utilization.

Aplicación de las redes neuronales artificiales para la estratificación de riesgo de mortalidad hospitalaria

Objetivo: Comparar la capacidad de prediccion de mortalidad hospitalaria de una red neuronal artificial (RNA) con el Acute Physiology and Chronic Health Evaluation II (APACHE II) y la regresion logistica (RL), y comparar la asignacion de probabilidades entre los distintos modelos. Metodo: Se recogen de forma prospectiva las variables necesarias para el calculo del APACHE II. Disponemos de 1.146 pacientes asignandose aleatoriamente (70 y 30%) al grupo de Desarrollo (800) y al de Validacion (346). Con las mismas variables se genera un modelo de RL y de RNA (perceptron 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 funcion de los criterios de discriminacion con el area bajo la curva ROC (ABC [IC del 95%]) y de calibracion 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 validacion, 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 asignacion de probabilidad entre RL y RN (8% del total) son pacientes con problemas neurologicos. Los peores resultados se obtienen en los pacientes traumaticos (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 significacion, ya que no trabaja con restricciones lineales ni de independencia de variables, con una diferente asignacion de probabilidad individual entre los modelos.

Transporte interhospitalario de largo recorrido. Utilidad de las escalas de gravedad

Objetivo Evaluar el riesgo de mortalidad hospitalaria del paciente trasladado desde un hospital comarcal a un hospital de referencia de segundo nivel mediante las escalas Rapid Acute Physiology Score (RAPS), Rapid Emergency Medicine Score (REMS), SAPS II y APACHE II. Diseno y ambito Estudio prospectivo observacional de los pacientes trasladados desde el hospital Sant Hospital de la Seu d’Urgell al Hospital Universitario Arnau de Vilanova de Lleida, a 132 km de distancia. Pacientes Se incluyo a 134 pacientes consecutivos trasladados desde octubre de 2005 a julio de 2007. Variables principales Se recogieron datos de filiacion, estancia, nivel de gravedad, diagnostico principal, servicio de destino y variables clinicas como ventilacion mecanica, uso de inotropicos, sedantes, relajantes musculares y antiarritmicos. Se considero como variable de resultado la mortalidad hospitalaria. Resultados La media ± desviacion estandar del tiempo de traslado fue 105 ± 14 min. El 31,6% ingresaron en una unidad de cuidados intensivos. Fallecieron durante el ingreso hospitalario 16 (11,9%) pacientes. El APACHE II y el SAPS II fueron significativamente mas elevados en los pacientes que fallecieron. El RAPS y el REMS no mostraron diferencias significativas entre fallecidos y supervivientes. A mayor puntuacion en APACHE II y SAPS II, se observo un aumento proporcional de mortalidad. El RAPS y el REMS no mostraron esta tendencia. El area bajo la curva ROC fue mejor para el APACHE II (0,76; intervalo de confianza [IC] del 95%, 0,63-0,89) y el SAPS II (0,78; IC del 95%, 0,67-0,89) que para el RAPS (0,59; IC del 95%, 0,43-0,75) y el REMS (0,63; IC del 95%, 0,49-0,78). Conclusiones El nivel de gravedad medido con APACHE II y SAPS II es un metodo util para determinar el pronostico de los pacientes trasladados permitiendo adecuar los recursos sanitarios fundamentalmente ante trayectos prolongados.

Clasificación de los trastornos mentales en la Unidad de Cuidados Intensivos

PURPOSE To determine the incidence and characteristics of mental disorders (MD) in the Intensive Care Unit (ICU), and to define a classification system adapted to the ICU environment. DESIGN A retrospective, descriptive analysis. SETTING Intensive Care Unit, Arnau de Vilanova Hospital in Lérida (Spain). PATIENTS All patients with MD admitted during 5-year period (January, 1 2004 to December 31, 2008). MAIN VARIABLES General variables included clinical-demographic data, diagnostic variables, procedures, severity score, length of stay and mortality. Specific variables included psychiatric history, screening for substance abuse, psychiatric assessment, monitoring and transfer to a psychiatric center. Classification of the MD was as follows: 1) acute substance intoxication (SI); 2) suicide attempts (SA); and 3) MD associated with the main diagnosis (AMD). RESULTS A total of 146 patients had MD (7.8%); they were predominantly male (74%) and were younger than the general ICU population (43.9 vs. 55.3 years, p<0.001). The ICU stays of the patients with MDs were shorter (4 days vs. 7 days, p<0.001), and there was less hospital mortality (17.1 vs. 25%, p<0.05). They also showed a higher incidence of pneumonia (19.9 vs. 13.8%, p<0.05), but no differences in the level of severity were observed. The SI group (24.7%) contained the highest number of young people; the SA group (36.3%) showed a predominance of women; and the AMD (39%) group had the longest stays and the highest mortality. Psychiatric consultation was carried out mainly in the SA group (62.3%). CONCLUSIONS MD is a relatively common problem in the ICU. Collaboration with the Psychiatry Department seldom occurs, but must be encouraged to develop fully integrated management of critical patients with MD.

Application of artificial neural networks for risk stratification of hospital mortality

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.

Cerebral salt wasting syndrome in a patient affected of spontaneous frontoparietal subdural haematoma

Ever since cerebral salt wasting syndrome (CSW) was first described in 1950, there have been debates over its existence and whether it has an important place in the differential diagnosis of hyponatraemia. We report the case of a neurosurgical patient with sustained hyponatraemia and abnormally high sodium loss in the urine, with signs of fluid volume depletion. Hyponatraemia was not corrected after an intravenous infusion of saline solution. Stable concentrations of blood sodium above 130 mmol/l were achieved with the administration of 100 mg of hydrocortisone daily, with an ensuing reduction in sodium elimination through the urine.

Gravedad en pacientes traumáticos ingresados en UCI. Modelos fisiológicos y anatómicos

INTRODUCTION The goals of this project were to compare both the anatomic and physiologic severity scores in trauma patients admitted to intensive care unit (ICU), and to elaborate mixed statistical models to improve the precision of the scores. METHODS A prospective study of cohorts. The combined medical/surgical ICU in a secondary university hospital. Seven hundred and eighty trauma patients admitted to ICU older than 16 years of age. Anatomic models (ISS and NISS) were compared and combined with physiological models (T-RTS, APACHE II [APII], and MPM II). The probability of death was calculated following the TRISS method. The discrimination was assessed using ROC curves (ABC [CI 95%]), and the calibration using the Hosmer-Lemeshoẃs H test. The mixed models were elaborated with the tree classification method type Chi Square Automatic Interaction Detection. RESULTS A 14% global mortality was recorded. The physiological models presented the best discrimination values (APII of 0.87 [0.84-0.90]). All models were affected by bad calibration (P<.01). The best mixed model resulted from the combination of APII and ISS (0.88 [0.83-0.90]). This model was able to differentiate between a 7.5% mortality for elderly patients with pathological antecedents and a 25% mortality in patients presenting traumatic brain injury, from a pool of patients with APII values ranging from 10 to 17 and an ISS threshold of 22. CONCLUSIONS The physiological models perform better than the anatomical models in traumatic patients admitted to the ICU. Patients with low scores in the physiological models require an anatomic analysis of the injuries to determine their severity.