Neuronal networks as predictors of death in pneumonia

Abstract

Background: the community-acquired pneumonia represents an important problem around the world. It is the fourth cause of death in our country. Prognoses indexes are helpful to early detect the high risk patients, but they have low sensibility and specificity.

Objective: to propose a predictive mathematical model of mortality by community- acquired pneumonia.

Materials and methods: longitudinal, analytic study in a universe of 73 patients and a non-probabilistic sample of 48. The Mann Whitney’s test was used to find variables with signification for mortality. Pearson correlation was applied to the significant variables and after that a mathematical model was elaborated and tested in a neuronal net created and trained for that. Later, data were introduced in a ROC curve to find the area under the curve as well as the coordinates of the cut-off point.

Results: the average age was 79 ±11 years and 50 % of the patients were women. Global mortality was around 27 %. The variables with behavioral differences were systolic arterial hypertension (x²=0.001), as well as the diastolic arterial pressure (x²=0.001). The creatinine value was (x²=0.03) and the respiratory frequency (x²=0.01). The oxygen pressure (x²=0.036), and also hemoglobin values and sodium (Na) level (x²=0.004) show a significant difference between groups and ages (x²=0.003) IC=0, 32.

Conclusions: this mathematical model is a useful tool at the patients´ bedside taking into account its help to clinical judgment when arriving to a more accurate prognosis.