Application of neural network model in early warning of nosocomial infection

  Objective  To explore the application of neural network model (NNM) in early warning of nosocomial infection (NI) for effective control of NI.

  Methods  We extrated data on 4 911 inpatients with infections (139 NI and 4 776 non-NI) in a terciary grade A hospital during March 2017. A algorithm formula was established using NNM combined with decision tree classifier after training based on the data collected. Then the established algorithm formula was atopted to predicate the occurence of infection inpatients in the hospital in a specific duration and compared the predictions to those of actual occurences to develop an optimal core algorithm for analysis of data from hospital infection information system.

  Results  For the models established, the fitting of classification tree was better than that of NNM and both the fitting of classification tree and NNM were vastly superior to that of logistic model. The predictive accuracy of classification tree model was better than that of logisitics model. The number of false negative prediction was obviously decreased with cross-complementing of coarsetree model and neuralnet model.

  Conclusion  In predication of NI occurence, the predictive accuracy of neural network model combined with classification tree model is obviously hgither than taht of conventional logistics model.