Time Series Analysis Of The Admission Of Patients With Acute Pancreatitis

Objective:1.To study the intrinsic regularity of the time series of hospitalization of patients with acute pancreatitis in our hospital.2.To compare the predictive efficacy for future hospitalization based on two modeling methods,the Autoregressive Moving Average Model and the Long-Short Term Memory neural network model.Method:To analyze the time series of hospitalized patients with acute pancreatitis from December 2006 to October 2018 in the Second Hospital of Hebei Medical University.This times series was divided into the first 123 months from December 2006 to February 2017 and the last 20 mouths from March 2017 to October 2018.1.R software was used to analyze the basic information of hospitalized patients during the total 143 mouths,such as the proportion of male and female,seasonal analysis,age distribution,etc.Moreover,in order to explore the potential relationship between air quality and the incidence of acute pancreatitis,the seasonal distribution of the air quality index and the nitrogen dioxide of Shijiazhuang City during this period of time is also illustrated.2.The hospitalization of acute pancreatitis in the first 123 months was utilized as a training set to predict the series of the next 20 months.The R-based ARIMA model and the LSTM neural network model were applied respectively to forecast.The index of the accuracy of the prediction is compared by calculating the root mean square error and the average absolute error value.3.The ARIMA model was also utilized to forecast the hospitalized patients with acute pancreatitis from November 2018 to June 2020.Based on the time series data from December 2006 to October 2018,the ARIMA modeling was generated by the stationarity test,difference,order determination,and fitted to optimized ARIMA model for forecasting.Finally,the inpatients with acute pancreatitis for the next 20 months were illustrated by the ARIMA model.Results:1.Firstly,the basic characteristics of inpatients with acute pancreatitis in 143 months was analyzed.The basic information of 6876 inpatients with acute pancreatitis were collected.4332 patients(63%)were male,while 2544 patients(37%)were female.The majority of patients were with ages between 30 to 50.84.1% of patients were admitted to our hospital for the first time.Based on seasonal analysis,the number of inpatients reached the peak in March,while the number fell to the bottom in December.In order to explore the potential association between air quality and incidence of acute pancreatits,the times series of AQI and NO2 were also analyzed.AQI and NO2 index peaked in December,January and February and rapidly declined in March.2.The prediction based on the ARIMA model trained by the first 123 months is with the same increasing tendency with the true values of last 20 months,but largely deviated from the true values.For the ARIMA,the RMSE value is 15.3375 and the MAE value is 11.64467.The prediction based on the LSTM neural network model are consistent with the true values,while the RMSE value is 3.891773 and the MAE value is 3.281802.The smaller RMSE and MAE in the prediction based on LSTM indicated higher accuracy in prediction.Based on the calculation of RMSE and MAE,and also based on the comparing between the predicted curves and the curves of true values,the LSTM neural network model has higher prediction accuracy and better prediction efficacy than the ARIMA model.However,interpretation by LSTM model is difficult,as too many parameters are trained by the neural network.While,the interpretation based on ARIMA is more viable,as only simplified parameters(p,d,q)are used to generate the model.Hence,3.The prediction for the inpatients with acute pancreatitis in the next 20 months was generated by the ARIMA model.It was predicted that the number of inpatients will continuously raising with a seasonal fluctuation.It was indicated that the number of hospitalizations was the highest in January and March and the lowest in November and December.Conclusion:1.The number of hospitalized patients with acute pancreatitis in our hospital is continuously increasing.In the seasonal analysis,the largest number of hospitalizations in March and the least in December.2.The LSTM neural network model has higher accuracy in terms of prediction for total tendency and seasonal fluctuation than the ARIMA model.However,poor interpretability and complicated settings of parameters limit its utility.3.Based on the ARIMA model,it was indicated that inpatients with acute pancreatitis will be gradually increasing in the next 20 months.The seasonality of inpatients in the future may be constant with pattern that shown in seasonal analysis.