| Objective:Sepsis is the acute systemic inflammatory response syndrome caused by infection, and it is one of the main causes of death in critically ill patients. The sepsis clinical manifestations are not specific and lack reliable criteria, so patients tend to deteriorate rapidly which leads to high mortality. By analyzing the early risk factors in the early stage of sepsis, the research identified factors associated with sepsis and established early warning model in order to validate the clinical value of the model.Methods:We collected 215 cases from January 2010 to October 2015, including 129 cases of infection,86 cases of sepsis. General condition, clinical symptoms and laboratory examination were recorded. Then we analyze related factors of infection causing sepsis. The logistic regression model with computer was used to construct the early warning model of sepsis by screening out the relevant factors. After that, we selected 30 severe infected cases and input data to verify the model, comparing with the 30 cases so as to analyze the clinical value of the early warning model of sepsis.Results:1) The general conditions of 129 infected patients and 86 sepsis patients revealed that the differences of clinical symptoms between two groups in hypoxemia, hypotension, disturbance of consciousness, and skin spots the were statistically significant. Sepsis group data was higher than those in infection group (P< 0.05). The differences of laboratory tests between two groups in platelet count, procalcitonin, total bilirubin, urea, creatinine, glucose and international normalized ratio were statistically significant. Platelet counts in sepsis group were less than those in infected group while the rest were higher than those in infection group (P< 0.05).2) Further logistic regression was used to analyze the related factors analyzed before. The results showed the sepsis early warning was associated with skin spots, platelet, procalcitonin, creatinine and international standard ratio(P< 0.05).After the establishment of logistic regression model to filter out these factors, we constructed computer software related to sepsis. The discriminant caused by the computer model was showed as follows: Z=1.893-3.788 (skin spots)-0.106 (PLT)-0.333 (PCT)+0.073 (Cr)+2.287 (INR). Therefore, the early warning probability was:P=1/(1+e-z). ROC curve analysis of the warning probability showed AUC=97.2%, and when sensitivity was 98.8%, the specificity was 74.8%.As a result, The Youden index was the biggest. At this time the forecast predictionP=7.6%, which could be thought as the best predictive accuracy.3) Artificial neural network model was used to verify the regression model, and the AUC=96.4%, which was similar with the AUC of logistic regerssion model.4) Sepsis related factors in 30 cases with severe infection were input into the computer model. Results of model were compared with the actual development of 30 cases. Results revealed 11 of the 30 cases developed sepsis actually. Sensitivity 45.5%, specificity 84.2%, positive predictive value 62.5%, negative predictive value 72.7%,and the diagnosis rate was 70.0% in early warning model. The datas of 11 cases before developing sepsis within 24h were analyzed in computer model showed 8 cases were sepsis. The 8 cases were the same as what they actually was and the sensitivity was 72.7%.Conclusions:1) Severe infection causing sepsis may be related to skin spots, platelet, procalcitonin, creatinineand international normalized ratio and so on.2) The study establishes the early warning model about sepsis and the calculate prediction P can determine whether the patient will develop sepsis.3) After validation and clinical application, this early warning model is of high sensitivity and specificity, with high accuracy by judging whether the infection will progress for sepsis. The model has clinical value. |
PDF Full Text Request