Prognostic model of acute pancreatitis

Abstract. The models for the prognosis of infectious complications and the outcome of acute pancreatitis are developed based on a mathematical analysis of the totality of clinical and laboratory data on the course of the disease, which have the form of a decision tree. It was revealed that laboratory indicators show statistically significant intergroup differences and allow to form a prognosis of the course of the disease. The threshold values of laboratory parameters calculated as a result of applying the classification and regression algorithm by constructing a decision tree are the nodal points for the distribution of patients according to the likelihood of further development of the disease. Thus, the presence of an international normalized ratio of more than 1,31 or an international normalized ratio of more than 1,31 and a hematocrit of less than 40% with a predictive probability of 80% is associated with the development of infectious complications in any period of the disease in the first 3 days of the development of the disease. If in the first 3 days of the disease the glucose level exceeds 11,55 mmol / L and the concentration of Ca²⁺ ions is less than 0,66 mmol/L, the probability of death is more than 70%. If the glucose level is more than 11,5 mmol/L and the level of Ca²⁺ ions is less than or equal to 0,66 mmol / L, or the glucose level is less than or equal to 11,5 mmol/L and the prothrombin index is less than or 83% and the hematocrit is less than or equal to 39,8% the probability of developing a fatal outcome at any period of the disease is 3 times higher compared to other patients. Prediction models of infectious complications and disease outcome have an accuracy of 78 and 87%, respectively. The use of these models allows stratification of patients upon admission to the hospital, highlighting the most disadvantaged patients in the prognostic plan. The models are quite simple and easy to use, do not require complex expensive studies. Thanks to the algorithm used in building models, they have the properties of self-learning, which in the future will increase their accuracy.