Yang Fei1, Kun Gao1, Wei-Qin Li2. 1. Medical School of Nanjing University, Jinling Hospital/Nanjing General Hospital of Nanjing Military Region of P.L.A., P.L.A. Research Institute of General Surgery, Nanjing, 210002, China. 2. Medical School of Nanjing University, Jinling Hospital/Nanjing General Hospital of Nanjing Military Region of P.L.A., P.L.A. Research Institute of General Surgery, Nanjing, 210002, China. Electronic address: chamskuler@163.com.
Abstract
OBJECTIVE: The aim of this study is to predict the risk of severe acute pancreatitis (SAP) associated with acute lung injury (ALI) by artificial neural networks (ANNs) model. METHODS: The ANNs and logistic regression model were constructed using clinical and laboratory data of 217 SAP patients. The models were first trained on 152 randomly chosen patients, validated and tested on the 33 patients and 32 patients respectively. Statistical indices were used to evaluate the value of the forecast in two models. RESULTS: The training set, validation set and test set were not significantly different for any of the 13 variables. After training, the back propagation network retained excellent pattern recognition ability. When the ANNs model was applied to the test set, it revealed a sensitivity of 87.5%, specificity of 83.3%. The accuracy was 84.43%. Significant differences could be found between ANNs model and logistic regression model in these parameter. When ANNs model was used to identify ALI, the area under receiver operating characteristic curve was 0.859 ± 0.048, which demonstrated the better overall properties than logistic regression modeling (AUC = 0.701 + 0.041) (95% CI: 0.664-0.857). Meanwhile, pancreatic necrosis rate, lactic dehydrogenase and oxyhemoglobin saturation were the important factors among all thirteen independent variable for ALI. CONCLUSION: The ANNs model was a valuable tool in dealing with the clinical risk prediction problem of ALI following to SAP. In addition, our approach can extract informative risk factors of ALI via the ANNs model.
OBJECTIVE: The aim of this study is to predict the risk of severe acute pancreatitis (SAP) associated with acute lung injury (ALI) by artificial neural networks (ANNs) model. METHODS: The ANNs and logistic regression model were constructed using clinical and laboratory data of 217 SAP patients. The models were first trained on 152 randomly chosen patients, validated and tested on the 33 patients and 32 patients respectively. Statistical indices were used to evaluate the value of the forecast in two models. RESULTS: The training set, validation set and test set were not significantly different for any of the 13 variables. After training, the back propagation network retained excellent pattern recognition ability. When the ANNs model was applied to the test set, it revealed a sensitivity of 87.5%, specificity of 83.3%. The accuracy was 84.43%. Significant differences could be found between ANNs model and logistic regression model in these parameter. When ANNs model was used to identify ALI, the area under receiver operating characteristic curve was 0.859 ± 0.048, which demonstrated the better overall properties than logistic regression modeling (AUC = 0.701 + 0.041) (95% CI: 0.664-0.857). Meanwhile, pancreatic necrosis rate, lactic dehydrogenase and oxyhemoglobin saturation were the important factors among all thirteen independent variable for ALI. CONCLUSION: The ANNs model was a valuable tool in dealing with the clinical risk prediction problem of ALI following to SAP. In addition, our approach can extract informative risk factors of ALI via the ANNs model.