Research On Mortality Prediction Analysis Based On ICU Patient Electronic Medical Record Data

With the rapid development of science and technology,computers play a pivotal role in many fields of society.2018 is the peak year of medical information development.In this year,a number of guidance documents have been issued to the hospital information system construction,starting from the traditional low-end data processing and turning to intelligent analysis.However,most hospitals currently cannot use data for medical decision-making,especially for how to use the historical data of past intensive care for future analysis.The main purpose of this paper is to use the computational physiology laboratory of the Massachusetts Institute of Technology and the multi-parameter intelligent intensive care database jointly developed by the Beth Israel Dikang Medical Center(BIDMC)and Philips Medical in the age group from 2001 to 2011.Data,improve ICU mortality prediction methods,and improve the effectiveness of mortality prediction models and the accuracy of prediction results.The 2012 ICU mortality was predicted by the improved BP neural network method proposed in this paper and the LC neural network model combined with the improved BP neural network model.The article mainly addresses the following three aspects:First of all,based on the MIMIC-Ⅲ database,there are some missing data in the diagnosis and treatment,because the data loss is very easy to cause the misjudgment of the death rate.How to fill the missing data becomes the fundamental problem of accurately determining the mortality.Based on this problem,this paper proposes an improved K-nearest neighbor algorithm on the missing data filling problem,and introduces the padded data into the pre-processing stage of mortality prediction to improve the accuracy of mortality prediction.Second,the Lee-Carter(LC)model has long been recognized as the preferred model for mortality prediction,and whether the model can achieve better results on the MIMIC-Ⅲ dataset.The experimental results show that the model has a large error between the prediction result and the actual value in some age ranges.Therefore,in order to reduce the error of the prediction result,this paper aims to avoid the problem that the BP neural network converges slowly and is easy to fall into the local optimum value.An improved BP neural network model was proposed to predict mortalityFinally,although the Lee-Carter model has certain limitations,it can be compensated for by other methods.Therefore,this paper proposes a linear weighted combination model of the model and the improved BP neural network model,which takes advantage of the LC model and the neural network,and improves the accuracy of the mortality prediction model.