Modelo preditivo para Diagnóstico da Sepse em Unidade de Terapia Intensiva

Abstract

The Intensive Care Unit (ICU) environments are specialized in assisting the patient severely ill. The development of these units drastically reduced the hospital mortality rate, however, the complex procedures on their patients favors the emergence of serious infections such as sepsis, which predisposes to death. The present study aimed to provide a probabilistic decision model for the early diagnosis of sepsis. To this end, we analyzed the data contained in the records of 100 patients hospitalized in a general ICU of a public hospital in the state of Paraíba, in the period from March to September 2011. The information was recorded in a proper instrument made by the researcher being studied variables: age, gender, initial diagnosis of the participants, the minimum and maximum axillary temperature, heart rate and respiratory rate, partial pressure of oxygen and carbon dioxide, serum lactate , potassium, sodium, total leukocyte count, segmented rods and, among others. We used binary logistic regression to determine the prediction model. Data analysis was performed using SPSS version 19.0. The findings showed that 63% of study participants were male, with a mean age of 62.5 years. Explanatory variables were considered: axillary temperature minimum, maximum axillary temperature, partial pressure of carbon dioxide, lactate, leukocyte count and the number of rods. ROC curve was possible to identify the optimal cutoff point to classify individuals according to the presence or absence of disease, which contributed to the making of the rule of decision for the early diagnosis of sepsis. Was performed to compare the degree of agreement between blood culture result considered the gold standard for diagnosis of infection and the model presented in this study using the Kappa coefficient, obtained a percentage of agreement of 0.93 is considered excellent. Despite the unprecedented nature of the research, proved to be early detection of sepsis with the adoption of statistical models as presented, however, further studies with different populations of ICUs should be performed in order to provide a better sample, making the results found reproducible in different clinical situations daily.