Predictive analysis of the number of hospitalized patients with acute pancreatitis based on time series model / 中华胰腺病杂志

ABSTRACT

Objective:To predict and analyze the number of acute pancreatitis (AP) inpatients based on time series model, and to explore the predictive efficiency of the model.Methods:Clinical data of AP inpatients in the Affiliated Hospital of Southwest Medical University from January 2014 to December 2019 were collected. R software was used to collect the time series of AP inpatients, and the trend and seasonal characteristics of AP inpatients from 2014 to 2018 were analyzed. Furthermore, the autoregressive moving average (ARIMA) model was established through stationarity test, model ordering and model testing steps, and the best selected model was used to predict the monthly number of inpatients in 2019 to verify its prediction efficiency.Results:A total of 3 939 AP patients were included in the study. The most common etiology for AP was cholestrogenic (48.2%), followed by hyperacylglyceremia (36.3%). The peak age of hospitalization was from 40 to 60 years old. Time series analysis showed that the number of AP inpatients increased year by year. The highest peak of the disease was from February to March, followed by September to November; and there was seasonal variation and the incidence was relatively small in summer. The established original training set sequence did not pass the stationarity test ( P=0.061), so the ARIMA model was established after it was transformed into a stationarity sequence by first-order difference. According to the criterion of minimum AIC value, ARIMA(2, 1, 1)(1, 1, 1) 12 was selected as the best model. The model was used to predict the number of AP inpatients in 2019, showing that it could better fit the trend of onset time and had good short-term prediction effect. The mean root error and absolute error were 6.8790 and 4.7783, respectively. Conclusions:The number of AP inpatients increases year by year with seasonal changes. ARIMA model is effective in predicting the number of AP inpatients and can be used for short-term prediction.